diff --git a/.vscode/settings.json b/.vscode/settings.json
new file mode 100644
index 000000000..3813ff8ac
--- /dev/null
+++ b/.vscode/settings.json
@@ -0,0 +1,5 @@
+{
+    "files.associations": {
+        "zmalloc.h": "c"
+    }
+}
\ No newline at end of file
diff --git a/src/Makefile b/src/Makefile
index adf32d557..680e3866b 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -119,7 +119,7 @@ ifeq ($(uname_S),DragonFly)
 else
 	# All the other OSes (notably Linux)
 	FINAL_LDFLAGS+= -rdynamic
-	FINAL_LIBS+=-ldl -pthread -lrt
+	FINAL_LIBS+=-ldl -pthread -lrt -lmemkind
 endif
 endif
 endif
@@ -164,11 +164,11 @@ endif
 
 REDIS_SERVER_NAME=redis-server
 REDIS_SENTINEL_NAME=redis-sentinel
-REDIS_SERVER_OBJ=adlist.o quicklist.o ae.o anet.o dict.o server.o sds.o zmalloc.o lzf_c.o lzf_d.o pqsort.o zipmap.o sha1.o ziplist.o release.o networking.o util.o object.o db.o replication.o rdb.o t_string.o t_list.o t_set.o t_zset.o t_hash.o config.o aof.o pubsub.o multi.o debug.o sort.o intset.o syncio.o cluster.o crc16.o endianconv.o slowlog.o scripting.o bio.o rio.o rand.o memtest.o crc64.o bitops.o sentinel.o notify.o setproctitle.o blocked.o hyperloglog.o latency.o sparkline.o redis-check-rdb.o redis-check-aof.o geo.o lazyfree.o module.o evict.o expire.o geohash.o geohash_helper.o childinfo.o defrag.o siphash.o rax.o t_stream.o listpack.o localtime.o lolwut.o lolwut5.o acl.o
+REDIS_SERVER_OBJ=adlist.o quicklist.o ae.o anet.o dict.o server.o sds.o zmalloc.o lzf_c.o lzf_d.o pqsort.o zipmap.o sha1.o ziplist.o release.o networking.o util.o object.o db.o replication.o rdb.o t_string.o t_list.o t_set.o t_zset.o t_hash.o config.o aof.o pubsub.o multi.o debug.o sort.o intset.o syncio.o cluster.o crc16.o endianconv.o slowlog.o scripting.o bio.o rio.o rand.o memtest.o crc64.o bitops.o sentinel.o notify.o setproctitle.o blocked.o hyperloglog.o latency.o sparkline.o redis-check-rdb.o redis-check-aof.o geo.o lazyfree.o module.o evict.o expire.o geohash.o geohash_helper.o childinfo.o defrag.o siphash.o rax.o t_stream.o listpack.o localtime.o lolwut.o lolwut5.o acl.o storage.o
 REDIS_CLI_NAME=redis-cli
-REDIS_CLI_OBJ=anet.o adlist.o dict.o redis-cli.o zmalloc.o release.o anet.o ae.o crc64.o siphash.o crc16.o
+REDIS_CLI_OBJ=anet.o adlist.o dict.o redis-cli.o zmalloc.o release.o anet.o ae.o crc64.o siphash.o crc16.o storage.o
 REDIS_BENCHMARK_NAME=redis-benchmark
-REDIS_BENCHMARK_OBJ=ae.o anet.o redis-benchmark.o adlist.o zmalloc.o redis-benchmark.o
+REDIS_BENCHMARK_OBJ=ae.o anet.o redis-benchmark.o adlist.o zmalloc.o redis-benchmark.o storage.o
 REDIS_CHECK_RDB_NAME=redis-check-rdb
 REDIS_CHECK_AOF_NAME=redis-check-aof
 
diff --git a/src/acl.c b/src/acl.c
index ecaae9268..b37b4719a 100644
--- a/src/acl.c
+++ b/src/acl.c
@@ -146,7 +146,7 @@ void ACLListFreeSds(void *item) {
  * If the user with such name already exists NULL is returned. */
 user *ACLCreateUser(const char *name, size_t namelen) {
     if (raxFind(Users,(unsigned char*)name,namelen) != raxNotFound) return NULL;
-    user *u = zmalloc(sizeof(*u));
+    user *u = zmalloc(sizeof(*u), MALLOC_LOCAL);
     u->name = sdsnewlen(name,namelen);
     u->flags = 0;
     u->allowed_subcommands = NULL;
@@ -267,7 +267,7 @@ void ACLAddAllowedSubcommand(user *u, unsigned long id, const char *sub) {
      * this user, we have to allocate the subcommands array. */
     if (u->allowed_subcommands == NULL) {
         u->allowed_subcommands = zcalloc(USER_COMMAND_BITS_COUNT *
-                                 sizeof(sds*));
+                                 sizeof(sds*), MALLOC_LOCAL);
     }
 
     /* We also need to enlarge the allocation pointing to the
diff --git a/src/adlist.c b/src/adlist.c
index ec5f8bbf4..359262521 100644
--- a/src/adlist.c
+++ b/src/adlist.c
@@ -42,7 +42,7 @@ list *listCreate(void)
 {
     struct list *list;
 
-    if ((list = zmalloc(sizeof(*list))) == NULL)
+    if ((list = zmalloc(sizeof(*list), MALLOC_SHARED)) == NULL)
         return NULL;
     list->head = list->tail = NULL;
     list->len = 0;
@@ -89,7 +89,7 @@ list *listAddNodeHead(list *list, void *value)
 {
     listNode *node;
 
-    if ((node = zmalloc(sizeof(*node))) == NULL)
+    if ((node = zmalloc(sizeof(*node), MALLOC_SHARED)) == NULL)
         return NULL;
     node->value = value;
     if (list->len == 0) {
@@ -115,7 +115,7 @@ list *listAddNodeTail(list *list, void *value)
 {
     listNode *node;
 
-    if ((node = zmalloc(sizeof(*node))) == NULL)
+    if ((node = zmalloc(sizeof(*node), MALLOC_SHARED)) == NULL)
         return NULL;
     node->value = value;
     if (list->len == 0) {
@@ -134,7 +134,7 @@ list *listAddNodeTail(list *list, void *value)
 list *listInsertNode(list *list, listNode *old_node, void *value, int after) {
     listNode *node;
 
-    if ((node = zmalloc(sizeof(*node))) == NULL)
+    if ((node = zmalloc(sizeof(*node), MALLOC_SHARED)) == NULL)
         return NULL;
     node->value = value;
     if (after) {
@@ -187,7 +187,7 @@ listIter *listGetIterator(list *list, int direction)
 {
     listIter *iter;
 
-    if ((iter = zmalloc(sizeof(*iter))) == NULL) return NULL;
+    if ((iter = zmalloc(sizeof(*iter), MALLOC_SHARED)) == NULL) return NULL;
     if (direction == AL_START_HEAD)
         iter->next = list->head;
     else
diff --git a/src/ae.c b/src/ae.c
index 53629ef77..7f59aecf1 100644
--- a/src/ae.c
+++ b/src/ae.c
@@ -64,9 +64,9 @@ aeEventLoop *aeCreateEventLoop(int setsize) {
     aeEventLoop *eventLoop;
     int i;
 
-    if ((eventLoop = zmalloc(sizeof(*eventLoop))) == NULL) goto err;
-    eventLoop->events = zmalloc(sizeof(aeFileEvent)*setsize);
-    eventLoop->fired = zmalloc(sizeof(aeFiredEvent)*setsize);
+    if ((eventLoop = zmalloc(sizeof(*eventLoop), MALLOC_LOCAL)) == NULL) goto err;
+    eventLoop->events = zmalloc(sizeof(aeFileEvent)*setsize, MALLOC_LOCAL);
+    eventLoop->fired = zmalloc(sizeof(aeFiredEvent)*setsize, MALLOC_LOCAL);
     if (eventLoop->events == NULL || eventLoop->fired == NULL) goto err;
     eventLoop->setsize = setsize;
     eventLoop->lastTime = time(NULL);
@@ -212,7 +212,7 @@ long long aeCreateTimeEvent(aeEventLoop *eventLoop, long long milliseconds,
     long long id = eventLoop->timeEventNextId++;
     aeTimeEvent *te;
 
-    te = zmalloc(sizeof(*te));
+    te = zmalloc(sizeof(*te), MALLOC_LOCAL);
     if (te == NULL) return AE_ERR;
     te->id = id;
     aeAddMillisecondsToNow(milliseconds,&te->when_sec,&te->when_ms);
diff --git a/src/ae_epoll.c b/src/ae_epoll.c
index 410aac70d..298bcf6ce 100644
--- a/src/ae_epoll.c
+++ b/src/ae_epoll.c
@@ -37,10 +37,10 @@ typedef struct aeApiState {
 } aeApiState;
 
 static int aeApiCreate(aeEventLoop *eventLoop) {
-    aeApiState *state = zmalloc(sizeof(aeApiState));
+    aeApiState *state = zmalloc(sizeof(aeApiState), MALLOC_LOCAL);
 
     if (!state) return -1;
-    state->events = zmalloc(sizeof(struct epoll_event)*eventLoop->setsize);
+    state->events = zmalloc(sizeof(struct epoll_event)*eventLoop->setsize, MALLOC_LOCAL);
     if (!state->events) {
         zfree(state);
         return -1;
diff --git a/src/ae_evport.c b/src/ae_evport.c
index 5c317becb..7b7fbe28e 100644
--- a/src/ae_evport.c
+++ b/src/ae_evport.c
@@ -74,7 +74,7 @@ typedef struct aeApiState {
 
 static int aeApiCreate(aeEventLoop *eventLoop) {
     int i;
-    aeApiState *state = zmalloc(sizeof(aeApiState));
+    aeApiState *state = zmalloc(sizeof(aeApiState), MALLOC_LOCAL);
     if (!state) return -1;
 
     state->portfd = port_create();
diff --git a/src/ae_kqueue.c b/src/ae_kqueue.c
index 6796f4ceb..36148bf58 100644
--- a/src/ae_kqueue.c
+++ b/src/ae_kqueue.c
@@ -42,7 +42,7 @@ static int aeApiCreate(aeEventLoop *eventLoop) {
     aeApiState *state = zmalloc(sizeof(aeApiState));
 
     if (!state) return -1;
-    state->events = zmalloc(sizeof(struct kevent)*eventLoop->setsize);
+    state->events = zmalloc(sizeof(struct kevent)*eventLoop->setsize, MALLOC_LOCAL);
     if (!state->events) {
         zfree(state);
         return -1;
diff --git a/src/ae_select.c b/src/ae_select.c
index c039a8ea3..7e6746aae 100644
--- a/src/ae_select.c
+++ b/src/ae_select.c
@@ -40,7 +40,7 @@ typedef struct aeApiState {
 } aeApiState;
 
 static int aeApiCreate(aeEventLoop *eventLoop) {
-    aeApiState *state = zmalloc(sizeof(aeApiState));
+    aeApiState *state = zmalloc(sizeof(aeApiState), MALLOC_LOCAL);
 
     if (!state) return -1;
     FD_ZERO(&state->rfds);
diff --git a/src/aof.c b/src/aof.c
index 46ae58324..ed9da54ef 100644
--- a/src/aof.c
+++ b/src/aof.c
@@ -143,7 +143,7 @@ void aofRewriteBufferAppend(unsigned char *s, unsigned long len) {
         if (len) { /* First block to allocate, or need another block. */
             int numblocks;
 
-            block = zmalloc(sizeof(*block));
+            block = zmalloc(sizeof(*block), MALLOC_LOCAL);
             block->free = AOF_RW_BUF_BLOCK_SIZE;
             block->used = 0;
             listAddNodeTail(server.aof_rewrite_buf_blocks,block);
@@ -627,7 +627,7 @@ void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv, int a
 /* In Redis commands are always executed in the context of a client, so in
  * order to load the append only file we need to create a fake client. */
 struct client *createFakeClient(void) {
-    struct client *c = zmalloc(sizeof(*c));
+    struct client *c = zmalloc(sizeof(*c), MALLOC_LOCAL);
 
     selectDb(c,0);
     c->fd = -1;
@@ -752,7 +752,7 @@ int loadAppendOnlyFile(char *filename) {
         argc = atoi(buf+1);
         if (argc < 1) goto fmterr;
 
-        argv = zmalloc(sizeof(robj*)*argc);
+        argv = zmalloc(sizeof(robj*)*argc, MALLOC_LOCAL);
         fakeClient->argc = argc;
         fakeClient->argv = argv;
 
diff --git a/src/bio.c b/src/bio.c
index 2af684570..13662197e 100644
--- a/src/bio.c
+++ b/src/bio.c
@@ -129,7 +129,7 @@ void bioInit(void) {
 }
 
 void bioCreateBackgroundJob(int type, void *arg1, void *arg2, void *arg3) {
-    struct bio_job *job = zmalloc(sizeof(*job));
+    struct bio_job *job = zmalloc(sizeof(*job), MALLOC_LOCAL);
 
     job->time = time(NULL);
     job->arg1 = arg1;
diff --git a/src/bitops.c b/src/bitops.c
index 8d03a7699..e2b53169a 100644
--- a/src/bitops.c
+++ b/src/bitops.c
@@ -621,9 +621,9 @@ void bitopCommand(client *c) {
 
     /* Lookup keys, and store pointers to the string objects into an array. */
     numkeys = c->argc - 3;
-    src = zmalloc(sizeof(unsigned char*) * numkeys);
-    len = zmalloc(sizeof(long) * numkeys);
-    objects = zmalloc(sizeof(robj*) * numkeys);
+    src = zmalloc(sizeof(unsigned char*) * numkeys, MALLOC_LOCAL);
+    len = zmalloc(sizeof(long) * numkeys, MALLOC_LOCAL);
+    objects = zmalloc(sizeof(robj*) * numkeys, MALLOC_LOCAL);
     for (j = 0; j < numkeys; j++) {
         o = lookupKeyRead(c->db,c->argv[j+3]);
         /* Handle non-existing keys as empty strings. */
diff --git a/src/blocked.c b/src/blocked.c
index f9e196626..24d674935 100644
--- a/src/blocked.c
+++ b/src/blocked.c
@@ -513,7 +513,7 @@ void blockForKeys(client *c, int btype, robj **keys, int numkeys, mstime_t timeo
          * is NULL for lists and sorted sets, or the stream ID for streams. */
         void *key_data = NULL;
         if (btype == BLOCKED_STREAM) {
-            key_data = zmalloc(sizeof(streamID));
+            key_data = zmalloc(sizeof(streamID), MALLOC_SHARED);
             memcpy(key_data,ids+j,sizeof(streamID));
         }
 
@@ -596,7 +596,7 @@ void signalKeyAsReady(redisDb *db, robj *key) {
     if (dictFind(db->ready_keys,key) != NULL) return;
 
     /* Ok, we need to queue this key into server.ready_keys. */
-    rl = zmalloc(sizeof(*rl));
+    rl = zmalloc(sizeof(*rl), MALLOC_SHARED);
     rl->key = key;
     rl->db = db;
     incrRefCount(key);
diff --git a/src/cluster.c b/src/cluster.c
index 1a3a348b5..bd04646d0 100644
--- a/src/cluster.c
+++ b/src/cluster.c
@@ -119,7 +119,7 @@ int clusterLoadConfig(char *filename) {
      *
      * To simplify we allocate 1024+CLUSTER_SLOTS*128 bytes per line. */
     maxline = 1024+CLUSTER_SLOTS*128;
-    line = zmalloc(maxline);
+    line = zmalloc(maxline, MALLOC_LOCAL);
     while(fgets(line,maxline,fp) != NULL) {
         int argc;
         sds *argv;
@@ -429,7 +429,7 @@ void clusterUpdateMyselfFlags(void) {
 void clusterInit(void) {
     int saveconf = 0;
 
-    server.cluster = zmalloc(sizeof(clusterState));
+    server.cluster = zmalloc(sizeof(clusterState), MALLOC_LOCAL);
     server.cluster->myself = NULL;
     server.cluster->currentEpoch = 0;
     server.cluster->state = CLUSTER_FAIL;
@@ -587,7 +587,7 @@ void clusterReset(int hard) {
  * -------------------------------------------------------------------------- */
 
 clusterLink *createClusterLink(clusterNode *node) {
-    clusterLink *link = zmalloc(sizeof(*link));
+    clusterLink *link = zmalloc(sizeof(*link), MALLOC_LOCAL);
     link->ctime = mstime();
     link->sndbuf = sdsempty();
     link->rcvbuf = sdsempty();
@@ -692,7 +692,7 @@ unsigned int keyHashSlot(char *key, int keylen) {
  * The node is created and returned to the user, but it is not automatically
  * added to the nodes hash table. */
 clusterNode *createClusterNode(char *nodename, int flags) {
-    clusterNode *node = zmalloc(sizeof(*node));
+    clusterNode *node = zmalloc(sizeof(*node), MALLOC_LOCAL);
 
     if (nodename)
         memcpy(node->name, nodename, CLUSTER_NAMELEN);
@@ -749,7 +749,7 @@ int clusterNodeAddFailureReport(clusterNode *failing, clusterNode *sender) {
     }
 
     /* Otherwise create a new report. */
-    fr = zmalloc(sizeof(*fr));
+    fr = zmalloc(sizeof(*fr), MALLOC_LOCAL);
     fr->node = sender;
     fr->time = mstime();
     listAddNodeTail(l,fr);
@@ -2401,7 +2401,7 @@ void clusterSendPing(clusterLink *link, int type) {
     /* Note: clusterBuildMessageHdr() expects the buffer to be always at least
      * sizeof(clusterMsg) or more. */
     if (totlen < (int)sizeof(clusterMsg)) totlen = sizeof(clusterMsg);
-    buf = zcalloc(totlen);
+    buf = zcalloc(totlen, MALLOC_LOCAL);
     hdr = (clusterMsg*) buf;
 
     /* Populate the header. */
@@ -2538,7 +2538,7 @@ void clusterSendPublish(clusterLink *link, robj *channel, robj *message) {
     if (totlen < sizeof(buf)) {
         payload = buf;
     } else {
-        payload = zmalloc(totlen);
+        payload = zmalloc(totlen, MALLOC_LOCAL);
         memcpy(payload,hdr,sizeof(*hdr));
         hdr = (clusterMsg*) payload;
     }
@@ -2607,7 +2607,7 @@ void clusterSendModule(clusterLink *link, uint64_t module_id, uint8_t type,
     if (totlen < sizeof(buf)) {
         heapbuf = buf;
     } else {
-        heapbuf = zmalloc(totlen);
+        heapbuf = zmalloc(totlen, MALLOC_LOCAL);
         memcpy(heapbuf,hdr,sizeof(*hdr));
         hdr = (clusterMsg*) heapbuf;
     }
@@ -4277,7 +4277,7 @@ NULL
         /* CLUSTER ADDSLOTS <slot> [slot] ... */
         /* CLUSTER DELSLOTS <slot> [slot] ... */
         int j, slot;
-        unsigned char *slots = zmalloc(CLUSTER_SLOTS);
+        unsigned char *slots = zmalloc(CLUSTER_SLOTS, MALLOC_LOCAL);
         int del = !strcasecmp(c->argv[1]->ptr,"delslots");
 
         memset(slots,0,CLUSTER_SLOTS);
@@ -4546,7 +4546,7 @@ NULL
         unsigned int keys_in_slot = countKeysInSlot(slot);
         if (maxkeys > keys_in_slot) maxkeys = keys_in_slot;
 
-        keys = zmalloc(sizeof(robj*)*maxkeys);
+        keys = zmalloc(sizeof(robj*)*maxkeys, MALLOC_LOCAL);
         numkeys = getKeysInSlot(slot, keys, maxkeys);
         addReplyArrayLen(c,numkeys);
         for (j = 0; j < numkeys; j++) {
@@ -5008,7 +5008,7 @@ migrateCachedSocket* migrateGetSocket(client *c, robj *host, robj *port, long ti
     }
 
     /* Add to the cache and return it to the caller. */
-    cs = zmalloc(sizeof(*cs));
+    cs = zmalloc(sizeof(*cs), MALLOC_LOCAL);
     cs->fd = fd;
     cs->last_dbid = -1;
     cs->last_use_time = server.unixtime;
@@ -5256,7 +5256,7 @@ try_again:
      * to propagate the MIGRATE as a DEL command (if no COPY option was given).
      * We allocate num_keys+1 because the additional argument is for "DEL"
      * command name itself. */
-    if (!copy) newargv = zmalloc(sizeof(robj*)*(num_keys+1));
+    if (!copy) newargv = zmalloc(sizeof(robj*)*(num_keys+1), MALLOC_LOCAL);
 
     for (j = 0; j < num_keys; j++) {
         if (syncReadLine(cs->fd, buf2, sizeof(buf2), timeout) <= 0) {
diff --git a/src/config.c b/src/config.c
index 91bbdee72..be1c45235 100644
--- a/src/config.c
+++ b/src/config.c
@@ -159,8 +159,8 @@ void queueLoadModule(sds path, sds *argv, int argc) {
     int i;
     struct moduleLoadQueueEntry *loadmod;
 
-    loadmod = zmalloc(sizeof(struct moduleLoadQueueEntry));
-    loadmod->argv = zmalloc(sizeof(robj*)*argc);
+    loadmod = zmalloc(sizeof(struct moduleLoadQueueEntry), MALLOC_LOCAL);
+    loadmod->argv = zmalloc(sizeof(robj*)*argc, MALLOC_LOCAL);
     loadmod->path = sdsnew(path);
     loadmod->argc = argc;
     for (i = 0; i < argc; i++) {
@@ -1672,7 +1672,7 @@ void rewriteConfigMarkAsProcessed(struct rewriteConfigState *state, const char *
  * If the old file does not exist at all, an empty state is returned. */
 struct rewriteConfigState *rewriteConfigReadOldFile(char *path) {
     FILE *fp = fopen(path,"r");
-    struct rewriteConfigState *state = zmalloc(sizeof(*state));
+    struct rewriteConfigState *state = zmalloc(sizeof(*state), MALLOC_LOCAL);
     char buf[CONFIG_MAX_LINE+1];
     int linenum = -1;
 
diff --git a/src/core.REMOVED.git-id b/src/core.REMOVED.git-id
new file mode 100644
index 000000000..251962f07
--- /dev/null
+++ b/src/core.REMOVED.git-id
@@ -0,0 +1 @@
+f6027990a0231d80395d15f8f867d3a31eea285d
\ No newline at end of file
diff --git a/src/db.c b/src/db.c
index f59ae8406..b44975b40 100644
--- a/src/db.c
+++ b/src/db.c
@@ -1200,7 +1200,7 @@ int *getKeysUsingCommandTable(struct redisCommand *cmd,robj **argv, int argc, in
 
     last = cmd->lastkey;
     if (last < 0) last = argc+last;
-    keys = zmalloc(sizeof(int)*((last - cmd->firstkey)+1));
+    keys = zmalloc(sizeof(int)*((last - cmd->firstkey)+1), MALLOC_SHARED);
     for (j = cmd->firstkey; j <= last; j += cmd->keystep) {
         if (j >= argc) {
             /* Modules commands, and standard commands with a not fixed number
@@ -1267,7 +1267,7 @@ int *zunionInterGetKeys(struct redisCommand *cmd, robj **argv, int argc, int *nu
     /* Keys in z{union,inter}store come from two places:
      * argv[1] = storage key,
      * argv[3...n] = keys to intersect */
-    keys = zmalloc(sizeof(int)*(num+1));
+    keys = zmalloc(sizeof(int)*(num+1), MALLOC_SHARED);
 
     /* Add all key positions for argv[3...n] to keys[] */
     for (i = 0; i < num; i++) keys[i] = 3+i;
@@ -1293,7 +1293,7 @@ int *evalGetKeys(struct redisCommand *cmd, robj **argv, int argc, int *numkeys)
         return NULL;
     }
 
-    keys = zmalloc(sizeof(int)*num);
+    keys = zmalloc(sizeof(int)*num, MALLOC_SHARED);
     *numkeys = num;
 
     /* Add all key positions for argv[3...n] to keys[] */
@@ -1314,7 +1314,7 @@ int *sortGetKeys(struct redisCommand *cmd, robj **argv, int argc, int *numkeys)
     UNUSED(cmd);
 
     num = 0;
-    keys = zmalloc(sizeof(int)*2); /* Alloc 2 places for the worst case. */
+    keys = zmalloc(sizeof(int)*2, MALLOC_SHARED); /* Alloc 2 places for the worst case. */
 
     keys[num++] = 1; /* <sort-key> is always present. */
 
@@ -1372,7 +1372,7 @@ int *migrateGetKeys(struct redisCommand *cmd, robj **argv, int argc, int *numkey
         }
     }
 
-    keys = zmalloc(sizeof(int)*num);
+    keys = zmalloc(sizeof(int)*num, MALLOC_SHARED);
     for (i = 0; i < num; i++) keys[i] = first+i;
     *numkeys = num;
     return keys;
@@ -1405,7 +1405,7 @@ int *georadiusGetKeys(struct redisCommand *cmd, robj **argv, int argc, int *numk
      * argv[1] = key,
      * argv[5...n] = stored key if present
      */
-    keys = zmalloc(sizeof(int) * num);
+    keys = zmalloc(sizeof(int) * num, MALLOC_SHARED);
 
     /* Add all key positions to keys[] */
     keys[0] = 1;
@@ -1454,7 +1454,7 @@ int *xreadGetKeys(struct redisCommand *cmd, robj **argv, int argc, int *numkeys)
     num /= 2; /* We have half the keys as there are arguments because
                  there are also the IDs, one per key. */
 
-    keys = zmalloc(sizeof(int) * num);
+    keys = zmalloc(sizeof(int) * num, MALLOC_SHARED);
     for (i = streams_pos+1; i < argc-num; i++) keys[i-streams_pos-1] = i;
     *numkeys = num;
     return keys;
@@ -1471,7 +1471,7 @@ void slotToKeyUpdateKey(robj *key, int add) {
     size_t keylen = sdslen(key->ptr);
 
     server.cluster->slots_keys_count[hashslot] += add ? 1 : -1;
-    if (keylen+2 > 64) indexed = zmalloc(keylen+2);
+    if (keylen+2 > 64) indexed = zmalloc(keylen+2, MALLOC_SHARED);
     indexed[0] = (hashslot >> 8) & 0xff;
     indexed[1] = hashslot & 0xff;
     memcpy(indexed+2,key->ptr,keylen);
diff --git a/src/debug.c b/src/debug.c
index a98bc61ad..d544d8832 100644
--- a/src/debug.c
+++ b/src/debug.c
@@ -345,7 +345,7 @@ NULL
         restartServer(flags,delay);
         addReplyError(c,"failed to restart the server. Check server logs.");
     } else if (!strcasecmp(c->argv[1]->ptr,"oom")) {
-        void *ptr = zmalloc(ULONG_MAX); /* Should trigger an out of memory. */
+        void *ptr = zmalloc(ULONG_MAX, MALLOC_LOCAL); /* Should trigger an out of memory. */
         zfree(ptr);
         addReply(c,shared.ok);
     } else if (!strcasecmp(c->argv[1]->ptr,"assert")) {
diff --git a/src/defrag.c b/src/defrag.c
index d67b6e253..7729c594c 100644
--- a/src/defrag.c
+++ b/src/defrag.c
@@ -72,8 +72,8 @@ void* activeDefragAlloc(void *ptr) {
     /* move this allocation to a new allocation.
      * make sure not to use the thread cache. so that we don't get back the same
      * pointers we try to free */
-    size = zmalloc_size(ptr);
-    newptr = zmalloc_no_tcache(size);
+    size = zmalloc_size(ptr, MALLOC_LOCAL);
+    newptr = zmalloc_no_tcache(size, MALLOC_LOCAL);
     memcpy(newptr, ptr, size);
     zfree_no_tcache(ptr);
     return newptr;
diff --git a/src/dict.c b/src/dict.c
index 2cf9d4839..3560eb3d3 100644
--- a/src/dict.c
+++ b/src/dict.c
@@ -111,7 +111,7 @@ static void _dictReset(dictht *ht)
 dict *dictCreate(dictType *type,
         void *privDataPtr)
 {
-    dict *d = zmalloc(sizeof(*d));
+    dict *d = zmalloc(sizeof(*d), MALLOC_SHARED);
 
     _dictInit(d,type,privDataPtr);
     return d;
@@ -160,7 +160,7 @@ int dictExpand(dict *d, unsigned long size)
     /* Allocate the new hash table and initialize all pointers to NULL */
     n.size = realsize;
     n.sizemask = realsize-1;
-    n.table = zcalloc(realsize*sizeof(dictEntry*));
+    n.table = zcalloc(realsize*sizeof(dictEntry*), MALLOC_SHARED);
     n.used = 0;
 
     /* Is this the first initialization? If so it's not really a rehashing
@@ -307,7 +307,7 @@ dictEntry *dictAddRaw(dict *d, void *key, dictEntry **existing)
      * system it is more likely that recently added entries are accessed
      * more frequently. */
     ht = dictIsRehashing(d) ? &d->ht[1] : &d->ht[0];
-    entry = zmalloc(sizeof(*entry));
+    entry = zmalloc(sizeof(*entry), MALLOC_SHARED);
     entry->next = ht->table[index];
     ht->table[index] = entry;
     ht->used++;
@@ -541,7 +541,7 @@ long long dictFingerprint(dict *d) {
 
 dictIterator *dictGetIterator(dict *d)
 {
-    dictIterator *iter = zmalloc(sizeof(*iter));
+    dictIterator *iter = zmalloc(sizeof(*iter), MALLOC_LOCAL);
 
     iter->d = d;
     iter->table = 0;
diff --git a/src/evict.c b/src/evict.c
index 773916ce8..cd033e62f 100644
--- a/src/evict.c
+++ b/src/evict.c
@@ -140,7 +140,7 @@ void evictionPoolAlloc(void) {
     struct evictionPoolEntry *ep;
     int j;
 
-    ep = zmalloc(sizeof(*ep)*EVPOOL_SIZE);
+    ep = zmalloc(sizeof(*ep)*EVPOOL_SIZE, MALLOC_LOCAL);
     for (j = 0; j < EVPOOL_SIZE; j++) {
         ep[j].idle = 0;
         ep[j].key = NULL;
diff --git a/src/geo.c b/src/geo.c
index 91a0421f5..1c792255e 100644
--- a/src/geo.c
+++ b/src/geo.c
@@ -51,7 +51,7 @@ int zslValueLteMax(double value, zrangespec *spec);
 
 /* Create a new array of geoPoints. */
 geoArray *geoArrayCreate(void) {
-    geoArray *ga = zmalloc(sizeof(*ga));
+    geoArray *ga = zmalloc(sizeof(*ga), MALLOC_SHARED);
     /* It gets allocated on first geoArrayAppend() call. */
     ga->array = NULL;
     ga->buckets = 0;
@@ -413,7 +413,7 @@ void geoaddCommand(client *c) {
 
     int elements = (c->argc - 2) / 3;
     int argc = 2+elements*2; /* ZADD key score ele ... */
-    robj **argv = zcalloc(argc*sizeof(robj*));
+    robj **argv = zcalloc(argc*sizeof(robj*), MALLOC_LOCAL);
     argv[0] = createRawStringObject("zadd",4);
     argv[1] = c->argv[1]; /* key */
     incrRefCount(argv[1]);
diff --git a/src/intset.c b/src/intset.c
index 4445a5ca6..4cf15f4db 100644
--- a/src/intset.c
+++ b/src/intset.c
@@ -95,7 +95,7 @@ static void _intsetSet(intset *is, int pos, int64_t value) {
 
 /* Create an empty intset. */
 intset *intsetNew(void) {
-    intset *is = zmalloc(sizeof(intset));
+    intset *is = zmalloc(sizeof(intset), MALLOC_SHARED);
     is->encoding = intrev32ifbe(INTSET_ENC_INT16);
     is->length = 0;
     return is;
diff --git a/src/latency.c b/src/latency.c
index 33aa1245b..5b4610993 100644
--- a/src/latency.c
+++ b/src/latency.c
@@ -102,7 +102,7 @@ void latencyAddSample(char *event, mstime_t latency) {
 
     /* Create the time series if it does not exist. */
     if (ts == NULL) {
-        ts = zmalloc(sizeof(*ts));
+        ts = zmalloc(sizeof(*ts), MALLOC_SHARED);
         ts->idx = 0;
         ts->max = 0;
         memset(ts->samples,0,sizeof(ts->samples));
diff --git a/src/listpack_malloc.h b/src/listpack_malloc.h
index 401ab6f74..4c0317528 100644
--- a/src/listpack_malloc.h
+++ b/src/listpack_malloc.h
@@ -39,7 +39,7 @@
 #ifndef LISTPACK_ALLOC_H
 #define LISTPACK_ALLOC_H
 #include "zmalloc.h"
-#define lp_malloc zmalloc
+#define lp_malloc(size) zmalloc(size, MALLOC_SHARED)
 #define lp_realloc zrealloc
 #define lp_free zfree
 #endif
diff --git a/src/lolwut5.c b/src/lolwut5.c
index 8408b378d..b247cfb13 100644
--- a/src/lolwut5.c
+++ b/src/lolwut5.c
@@ -71,10 +71,10 @@ void lwTranslatePixelsGroup(int byte, char *output) {
 
 /* Allocate and return a new canvas of the specified size. */
 lwCanvas *lwCreateCanvas(int width, int height) {
-    lwCanvas *canvas = zmalloc(sizeof(*canvas));
+    lwCanvas *canvas = zmalloc(sizeof(*canvas), MALLOC_SHARED);
     canvas->width = width;
     canvas->height = height;
-    canvas->pixels = zmalloc(width*height);
+    canvas->pixels = zmalloc(width*height, MALLOC_SHARED);
     memset(canvas->pixels,0,width*height);
     return canvas;
 }
diff --git a/src/module.c b/src/module.c
index 8954fcdf0..b06493a1b 100644
--- a/src/module.c
+++ b/src/module.c
@@ -280,7 +280,7 @@ void RM_FreeDict(RedisModuleCtx *ctx, RedisModuleDict *d);
  * and in general is taken into account as memory allocated by Redis.
  * You should avoid using malloc(). */
 void *RM_Alloc(size_t bytes) {
-    return zmalloc(bytes);
+    return zmalloc(bytes, MALLOC_LOCAL);
 }
 
 /* Use like calloc(). Memory allocated with this function is reported in
@@ -288,7 +288,7 @@ void *RM_Alloc(size_t bytes) {
  * and in general is taken into account as memory allocated by Redis.
  * You should avoid using calloc() directly. */
 void *RM_Calloc(size_t nmemb, size_t size) {
-    return zcalloc(nmemb*size);
+    return zcalloc(nmemb*size, MALLOC_LOCAL);
 }
 
 /* Use like realloc() for memory obtained with RedisModule_Alloc(). */
@@ -354,7 +354,7 @@ void *RM_PoolAlloc(RedisModuleCtx *ctx, size_t bytes) {
     if (left < bytes) {
         size_t blocksize = REDISMODULE_POOL_ALLOC_MIN_SIZE;
         if (blocksize < bytes) blocksize = bytes;
-        b = zmalloc(sizeof(*b) + blocksize);
+        b = zmalloc(sizeof(*b) + blocksize, MALLOC_LOCAL);
         b->size = blocksize;
         b->used = 0;
         b->next = ctx->pa_head;
@@ -668,10 +668,10 @@ int RM_CreateCommand(RedisModuleCtx *ctx, const char *name, RedisModuleCmdFunc c
      *
      * Note that we use the Redis command table 'getkeys_proc' in order to
      * pass a reference to the command proxy structure. */
-    cp = zmalloc(sizeof(*cp));
+    cp = zmalloc(sizeof(*cp), MALLOC_LOCAL);
     cp->module = ctx->module;
     cp->func = cmdfunc;
-    cp->rediscmd = zmalloc(sizeof(*rediscmd));
+    cp->rediscmd = zmalloc(sizeof(*rediscmd), MALLOC_LOCAL);
     cp->rediscmd->name = cmdname;
     cp->rediscmd->proc = RedisModuleCommandDispatcher;
     cp->rediscmd->arity = -1;
@@ -695,7 +695,7 @@ void RM_SetModuleAttribs(RedisModuleCtx *ctx, const char *name, int ver, int api
     RedisModule *module;
 
     if (ctx->module != NULL) return;
-    module = zmalloc(sizeof(*module));
+    module = zmalloc(sizeof(*module), MALLOC_LOCAL);
     module->name = sdsnew((char*)name);
     module->ver = ver;
     module->apiver = apiver;
@@ -1470,7 +1470,7 @@ void *RM_OpenKey(RedisModuleCtx *ctx, robj *keyname, int mode) {
     }
 
     /* Setup the key handle. */
-    kp = zmalloc(sizeof(*kp));
+    kp = zmalloc(sizeof(*kp), MALLOC_LOCAL);
     kp->ctx = ctx;
     kp->db = ctx->client->db;
     kp->key = keyname;
@@ -2392,7 +2392,7 @@ int RM_HashGet(RedisModuleKey *key, int flags, ...) {
  * is processed as needed. Initially we just make sure to set the right
  * reply type, which is extremely cheap to do. */
 RedisModuleCallReply *moduleCreateCallReplyFromProto(RedisModuleCtx *ctx, sds proto) {
-    RedisModuleCallReply *reply = zmalloc(sizeof(*reply));
+    RedisModuleCallReply *reply = zmalloc(sizeof(*reply), MALLOC_LOCAL);
     reply->ctx = ctx;
     reply->proto = proto;
     reply->protolen = sdslen(proto);
@@ -2482,7 +2482,7 @@ void moduleParseCallReply_Array(RedisModuleCallReply *reply) {
         return;
     }
 
-    reply->val.array = zmalloc(sizeof(RedisModuleCallReply)*arraylen);
+    reply->val.array = zmalloc(sizeof(RedisModuleCallReply)*arraylen, MALLOC_LOCAL);
     reply->len = arraylen;
     for (j = 0; j < arraylen; j++) {
         RedisModuleCallReply *ele = reply->val.array+j;
@@ -2996,7 +2996,7 @@ moduleType *RM_CreateDataType(RedisModuleCtx *ctx, const char *name, int encver,
         moduleTypeFreeFunc free;
     } *tms = (struct typemethods*) typemethods_ptr;
 
-    moduleType *mt = zcalloc(sizeof(*mt));
+    moduleType *mt = zcalloc(sizeof(*mt), MALLOC_LOCAL);
     mt->id = id;
     mt->module = ctx->module;
     mt->rdb_load = tms->rdb_load;
@@ -3393,7 +3393,7 @@ void RM_EmitAOF(RedisModuleIO *io, const char *cmdname, const char *fmt, ...) {
 RedisModuleCtx *RM_GetContextFromIO(RedisModuleIO *io) {
     if (io->ctx) return io->ctx; /* Can't have more than one... */
     RedisModuleCtx ctxtemplate = REDISMODULE_CTX_INIT;
-    io->ctx = zmalloc(sizeof(RedisModuleCtx));
+    io->ctx = zmalloc(sizeof(RedisModuleCtx), MALLOC_LOCAL);
     *(io->ctx) = ctxtemplate;
     io->ctx->module = io->type->module;
     io->ctx->client = NULL;
@@ -3530,7 +3530,7 @@ RedisModuleBlockedClient *RM_BlockClient(RedisModuleCtx *ctx, RedisModuleCmdFunc
     int islua = c->flags & CLIENT_LUA;
     int ismulti = c->flags & CLIENT_MULTI;
 
-    c->bpop.module_blocked_handle = zmalloc(sizeof(RedisModuleBlockedClient));
+    c->bpop.module_blocked_handle = zmalloc(sizeof(RedisModuleBlockedClient), MALLOC_LOCAL);
     RedisModuleBlockedClient *bc = c->bpop.module_blocked_handle;
 
     /* We need to handle the invalid operation of calling modules blocking
@@ -3780,7 +3780,7 @@ int RM_BlockedClientDisconnected(RedisModuleCtx *ctx) {
  * TODO: thread safe contexts do not inherit the blocked client
  * selected database. */
 RedisModuleCtx *RM_GetThreadSafeContext(RedisModuleBlockedClient *bc) {
-    RedisModuleCtx *ctx = zmalloc(sizeof(*ctx));
+    RedisModuleCtx *ctx = zmalloc(sizeof(*ctx), MALLOC_LOCAL);
     RedisModuleCtx empty = REDISMODULE_CTX_INIT;
     memcpy(ctx,&empty,sizeof(empty));
     if (bc) {
@@ -3880,7 +3880,7 @@ void moduleReleaseGIL(void) {
  * See https://redis.io/topics/notifications for more information.
  */
 int RM_SubscribeToKeyspaceEvents(RedisModuleCtx *ctx, int types, RedisModuleNotificationFunc callback) {
-    RedisModuleKeyspaceSubscriber *sub = zmalloc(sizeof(*sub));
+    RedisModuleKeyspaceSubscriber *sub = zmalloc(sizeof(*sub), MALLOC_LOCAL);
     sub->module = ctx->module;
     sub->event_mask = types;
     sub->notify_callback = callback;
@@ -4017,7 +4017,7 @@ void RM_RegisterClusterMessageReceiver(RedisModuleCtx *ctx, uint8_t type, RedisM
 
     /* Not found, let's add it. */
     if (callback) {
-        r = zmalloc(sizeof(*r));
+        r = zmalloc(sizeof(*r), MALLOC_LOCAL);
         r->module_id = module_id;
         r->module = ctx->module;
         r->callback = callback;
@@ -4069,14 +4069,14 @@ char **RM_GetClusterNodesList(RedisModuleCtx *ctx, size_t *numnodes) {
 
     if (!server.cluster_enabled) return NULL;
     size_t count = dictSize(server.cluster->nodes);
-    char **ids = zmalloc((count+1)*REDISMODULE_NODE_ID_LEN);
+    char **ids = zmalloc((count+1)*REDISMODULE_NODE_ID_LEN, MALLOC_LOCAL);
     dictIterator *di = dictGetIterator(server.cluster->nodes);
     dictEntry *de;
     int j = 0;
     while((de = dictNext(di)) != NULL) {
         clusterNode *node = dictGetVal(de);
         if (node->flags & (CLUSTER_NODE_NOADDR|CLUSTER_NODE_HANDSHAKE)) continue;
-        ids[j] = zmalloc(REDISMODULE_NODE_ID_LEN);
+        ids[j] = zmalloc(REDISMODULE_NODE_ID_LEN, MALLOC_LOCAL);
         memcpy(ids[j],node->name,REDISMODULE_NODE_ID_LEN);
         j++;
     }
@@ -4269,7 +4269,7 @@ int moduleTimerHandler(struct aeEventLoop *eventLoop, long long id, void *client
  * the specified function using `data` as argument. The returned timer ID can be
  * used to get information from the timer or to stop it before it fires. */
 RedisModuleTimerID RM_CreateTimer(RedisModuleCtx *ctx, mstime_t period, RedisModuleTimerProc callback, void *data) {
-    RedisModuleTimer *timer = zmalloc(sizeof(*timer));
+    RedisModuleTimer *timer = zmalloc(sizeof(*timer), MALLOC_LOCAL);
     timer->module = ctx->module;
     timer->callback = callback;
     timer->data = data;
@@ -4367,7 +4367,7 @@ int RM_GetTimerInfo(RedisModuleCtx *ctx, RedisModuleTimerID id, uint64_t *remain
  *    Next / Prev dictionary iterator calls.
  */
 RedisModuleDict *RM_CreateDict(RedisModuleCtx *ctx) {
-    struct RedisModuleDict *d = zmalloc(sizeof(*d));
+    struct RedisModuleDict *d = zmalloc(sizeof(*d), MALLOC_LOCAL);
     d->rax = raxNew();
     if (ctx != NULL) autoMemoryAdd(ctx,REDISMODULE_AM_DICT,d);
     return d;
@@ -4467,7 +4467,7 @@ int RM_DictDel(RedisModuleDict *d, RedisModuleString *key, void *oldval) {
  * REDISMODULE_ERR at the first call, otherwise they'll produce elements.
  */
 RedisModuleDictIter *RM_DictIteratorStartC(RedisModuleDict *d, const char *op, void *key, size_t keylen) {
-    RedisModuleDictIter *di = zmalloc(sizeof(*di));
+    RedisModuleDictIter *di = zmalloc(sizeof(*di), MALLOC_LOCAL);
     di->dict = d;
     raxStart(&di->ri,d->rax);
     raxSeek(&di->ri,op,key,keylen);
diff --git a/src/multi.c b/src/multi.c
index 71090d8ed..720ec90e2 100644
--- a/src/multi.c
+++ b/src/multi.c
@@ -63,7 +63,7 @@ void queueMultiCommand(client *c) {
     mc = c->mstate.commands+c->mstate.count;
     mc->cmd = c->cmd;
     mc->argc = c->argc;
-    mc->argv = zmalloc(sizeof(robj*)*c->argc);
+    mc->argv = zmalloc(sizeof(robj*)*c->argc, MALLOC_LOCAL);
     memcpy(mc->argv,c->argv,sizeof(robj*)*c->argc);
     for (j = 0; j < c->argc; j++)
         incrRefCount(mc->argv[j]);
@@ -253,7 +253,7 @@ void watchForKey(client *c, robj *key) {
     }
     listAddNodeTail(clients,c);
     /* Add the new key to the list of keys watched by this client */
-    wk = zmalloc(sizeof(*wk));
+    wk = zmalloc(sizeof(*wk), MALLOC_SHARED);
     wk->key = key;
     wk->db = c->db;
     incrRefCount(key);
diff --git a/src/networking.c b/src/networking.c
index 9e62cb6be..b853da5bd 100644
--- a/src/networking.c
+++ b/src/networking.c
@@ -57,7 +57,7 @@ size_t getStringObjectSdsUsedMemory(robj *o) {
 /* Client.reply list dup and free methods. */
 void *dupClientReplyValue(void *o) {
     clientReplyBlock *old = o;
-    clientReplyBlock *buf = zmalloc(sizeof(clientReplyBlock) + old->size);
+    clientReplyBlock *buf = zmalloc(sizeof(clientReplyBlock) + old->size, MALLOC_LOCAL);
     memcpy(buf, o, sizeof(clientReplyBlock) + old->size);
     return buf;
 }
@@ -83,7 +83,7 @@ void linkClient(client *c) {
 }
 
 client *createClient(int fd) {
-    client *c = zmalloc(sizeof(client));
+    client *c = zmalloc(sizeof(client), MALLOC_LOCAL);
 
     /* passing -1 as fd it is possible to create a non connected client.
      * This is useful since all the commands needs to be executed
@@ -281,7 +281,7 @@ void _addReplyProtoToList(client *c, const char *s, size_t len) {
         /* Create a new node, make sure it is allocated to at
          * least PROTO_REPLY_CHUNK_BYTES */
         size_t size = len < PROTO_REPLY_CHUNK_BYTES? PROTO_REPLY_CHUNK_BYTES: len;
-        tail = zmalloc(size + sizeof(clientReplyBlock));
+        tail = zmalloc(size + sizeof(clientReplyBlock), MALLOC_LOCAL);
         /* take over the allocation's internal fragmentation */
         tail->size = zmalloc_usable(tail) - sizeof(clientReplyBlock);
         tail->used = len;
@@ -460,7 +460,7 @@ void setDeferredAggregateLen(client *c, void *node, long length, char prefix) {
         listDelNode(c->reply,ln);
     } else {
         /* Create a new node */
-        clientReplyBlock *buf = zmalloc(lenstr_len + sizeof(clientReplyBlock));
+        clientReplyBlock *buf = zmalloc(lenstr_len + sizeof(clientReplyBlock), MALLOC_LOCAL);
         /* Take over the allocation's internal fragmentation */
         buf->size = zmalloc_usable(buf) - sizeof(clientReplyBlock);
         buf->used = lenstr_len;
@@ -1333,7 +1333,7 @@ int processInlineBuffer(client *c) {
     /* Setup argv array on client structure */
     if (argc) {
         if (c->argv) zfree(c->argv);
-        c->argv = zmalloc(sizeof(robj*)*argc);
+        c->argv = zmalloc(sizeof(robj*)*argc, MALLOC_LOCAL);
     }
 
     /* Create redis objects for all arguments. */
@@ -1345,7 +1345,7 @@ int processInlineBuffer(client *c) {
             sdsfree(argv[j]);
         }
     }
-    zfree(argv);
+    sds_free(argv);
     return C_OK;
 }
 
@@ -1431,7 +1431,7 @@ int processMultibulkBuffer(client *c) {
 
         /* Setup argv array on client structure */
         if (c->argv) zfree(c->argv);
-        c->argv = zmalloc(sizeof(robj*)*c->multibulklen);
+        c->argv = zmalloc(sizeof(robj*)*c->multibulklen, MALLOC_LOCAL);
     }
 
     serverAssertWithInfo(c,NULL,c->multibulklen > 0);
@@ -2123,7 +2123,7 @@ void rewriteClientCommandVector(client *c, int argc, ...) {
     int j;
     robj **argv; /* The new argument vector */
 
-    argv = zmalloc(sizeof(robj*)*argc);
+    argv = zmalloc(sizeof(robj*)*argc, MALLOC_LOCAL);
     va_start(ap,argc);
     for (j = 0; j < argc; j++) {
         robj *a;
diff --git a/src/object.c b/src/object.c
index ec0bd02ee..ddd50b553 100644
--- a/src/object.c
+++ b/src/object.c
@@ -39,7 +39,7 @@
 /* ===================== Creation and parsing of objects ==================== */
 
 robj *createObject(int type, void *ptr) {
-    robj *o = zmalloc(sizeof(*o));
+    robj *o = salloc_obj(); //zmalloc(sizeof(*o), MALLOC_SHARED);
     o->type = type;
     o->encoding = OBJ_ENCODING_RAW;
     o->ptr = ptr;
@@ -82,7 +82,7 @@ robj *createRawStringObject(const char *ptr, size_t len) {
  * an object where the sds string is actually an unmodifiable string
  * allocated in the same chunk as the object itself. */
 robj *createEmbeddedStringObject(const char *ptr, size_t len) {
-    robj *o = zmalloc(sizeof(robj)+sizeof(struct sdshdr8)+len+1);
+    robj *o = zmalloc(sizeof(robj)+sizeof(struct sdshdr8)+len+1, MALLOC_SHARED);
     struct sdshdr8 *sh = (void*)(o+1);
 
     o->type = OBJ_STRING;
@@ -247,7 +247,7 @@ robj *createHashObject(void) {
 }
 
 robj *createZsetObject(void) {
-    zset *zs = zmalloc(sizeof(*zs));
+    zset *zs = zmalloc(sizeof(*zs), MALLOC_SHARED);
     robj *o;
 
     zs->dict = dictCreate(&zsetDictType,NULL);
@@ -272,7 +272,7 @@ robj *createStreamObject(void) {
 }
 
 robj *createModuleObject(moduleType *mt, void *value) {
-    moduleValue *mv = zmalloc(sizeof(*mv));
+    moduleValue *mv = zmalloc(sizeof(*mv), MALLOC_SHARED);
     mv->type = mt;
     mv->value = value;
     return createObject(OBJ_MODULE,mv);
@@ -362,7 +362,10 @@ void decrRefCount(robj *o) {
         case OBJ_STREAM: freeStreamObject(o); break;
         default: serverPanic("Unknown object type"); break;
         }
-        zfree(o);
+        if (o->type == OBJ_STRING && o->encoding == OBJ_ENCODING_EMBSTR)
+            zfree(o);
+        else
+            sfree_obj(o);
     } else {
         if (o->refcount <= 0) serverPanic("decrRefCount against refcount <= 0");
         if (o->refcount != OBJ_SHARED_REFCOUNT) o->refcount--;
@@ -945,7 +948,7 @@ struct redisMemOverhead *getMemoryOverheadData(void) {
     size_t mem_total = 0;
     size_t mem = 0;
     size_t zmalloc_used = zmalloc_used_memory();
-    struct redisMemOverhead *mh = zcalloc(sizeof(*mh));
+    struct redisMemOverhead *mh = zcalloc(sizeof(*mh), MALLOC_LOCAL);
 
     mh->total_allocated = zmalloc_used;
     mh->startup_allocated = server.initial_memory_usage;
diff --git a/src/pubsub.c b/src/pubsub.c
index 994dd9734..a7ee67400 100644
--- a/src/pubsub.c
+++ b/src/pubsub.c
@@ -209,7 +209,7 @@ int pubsubSubscribePattern(client *c, robj *pattern) {
         pubsubPattern *pat;
         listAddNodeTail(c->pubsub_patterns,pattern);
         incrRefCount(pattern);
-        pat = zmalloc(sizeof(*pat));
+        pat = zmalloc(sizeof(*pat), MALLOC_LOCAL);
         pat->pattern = getDecodedObject(pattern);
         pat->client = c;
         listAddNodeTail(server.pubsub_patterns,pat);
diff --git a/src/quicklist.c b/src/quicklist.c
index 7b5484116..ea0a2ae77 100644
--- a/src/quicklist.c
+++ b/src/quicklist.c
@@ -94,7 +94,7 @@ static const size_t optimization_level[] = {4096, 8192, 16384, 32768, 65536};
 quicklist *quicklistCreate(void) {
     struct quicklist *quicklist;
 
-    quicklist = zmalloc(sizeof(*quicklist));
+    quicklist = zmalloc(sizeof(*quicklist), MALLOC_SHARED);
     quicklist->head = quicklist->tail = NULL;
     quicklist->len = 0;
     quicklist->count = 0;
@@ -137,7 +137,7 @@ quicklist *quicklistNew(int fill, int compress) {
 
 REDIS_STATIC quicklistNode *quicklistCreateNode(void) {
     quicklistNode *node;
-    node = zmalloc(sizeof(*node));
+    node = zmalloc(sizeof(*node), MALLOC_SHARED);
     node->zl = NULL;
     node->count = 0;
     node->sz = 0;
@@ -184,7 +184,7 @@ REDIS_STATIC int __quicklistCompressNode(quicklistNode *node) {
     if (node->sz < MIN_COMPRESS_BYTES)
         return 0;
 
-    quicklistLZF *lzf = zmalloc(sizeof(*lzf) + node->sz);
+    quicklistLZF *lzf = zmalloc(sizeof(*lzf) + node->sz, MALLOC_SHARED);
 
     /* Cancel if compression fails or doesn't compress small enough */
     if (((lzf->sz = lzf_compress(node->zl, node->sz, lzf->compressed,
@@ -217,7 +217,7 @@ REDIS_STATIC int __quicklistDecompressNode(quicklistNode *node) {
     node->attempted_compress = 0;
 #endif
 
-    void *decompressed = zmalloc(node->sz);
+    void *decompressed = zmalloc(node->sz, MALLOC_SHARED);
     quicklistLZF *lzf = (quicklistLZF *)node->zl;
     if (lzf_decompress(lzf->compressed, lzf->sz, decompressed, node->sz) == 0) {
         /* Someone requested decompress, but we can't decompress.  Not good. */
@@ -799,7 +799,7 @@ REDIS_STATIC quicklistNode *_quicklistSplitNode(quicklistNode *node, int offset,
     size_t zl_sz = node->sz;
 
     quicklistNode *new_node = quicklistCreateNode();
-    new_node->zl = zmalloc(zl_sz);
+    new_node->zl = zmalloc(zl_sz, MALLOC_SHARED);
 
     /* Copy original ziplist so we can split it */
     memcpy(new_node->zl, node->zl, zl_sz);
@@ -1048,7 +1048,7 @@ int quicklistCompare(unsigned char *p1, unsigned char *p2, int p2_len) {
 quicklistIter *quicklistGetIterator(const quicklist *quicklist, int direction) {
     quicklistIter *iter;
 
-    iter = zmalloc(sizeof(*iter));
+    iter = zmalloc(sizeof(*iter), MALLOC_LOCAL);
 
     if (direction == AL_START_HEAD) {
         iter->current = quicklist->head;
@@ -1195,10 +1195,10 @@ quicklist *quicklistDup(quicklist *orig) {
         if (current->encoding == QUICKLIST_NODE_ENCODING_LZF) {
             quicklistLZF *lzf = (quicklistLZF *)current->zl;
             size_t lzf_sz = sizeof(*lzf) + lzf->sz;
-            node->zl = zmalloc(lzf_sz);
+            node->zl = zmalloc(lzf_sz, MALLOC_SHARED);
             memcpy(node->zl, current->zl, lzf_sz);
         } else if (current->encoding == QUICKLIST_NODE_ENCODING_RAW) {
-            node->zl = zmalloc(current->sz);
+            node->zl = zmalloc(current->sz, MALLOC_SHARED);
             memcpy(node->zl, current->zl, current->sz);
         }
 
@@ -1372,7 +1372,7 @@ int quicklistPopCustom(quicklist *quicklist, int where, unsigned char **data,
 REDIS_STATIC void *_quicklistSaver(unsigned char *data, unsigned int sz) {
     unsigned char *vstr;
     if (data) {
-        vstr = zmalloc(sz);
+        vstr = zmalloc(sz, MALLOC_SHARED);
         memcpy(vstr, data, sz);
         return vstr;
     }
diff --git a/src/rax_malloc.h b/src/rax_malloc.h
index 9295985c6..a5aea44d9 100644
--- a/src/rax_malloc.h
+++ b/src/rax_malloc.h
@@ -38,7 +38,7 @@
 #ifndef RAX_ALLOC_H
 #define RAX_ALLOC_H
 #include "zmalloc.h"
-#define rax_malloc zmalloc
+#define rax_malloc(size) zmalloc(size, MALLOC_SHARED)
 #define rax_realloc zrealloc
 #define rax_free zfree
 #endif
diff --git a/src/rdb.c b/src/rdb.c
index 351b90774..32a3df1d5 100644
--- a/src/rdb.c
+++ b/src/rdb.c
@@ -32,6 +32,7 @@
 #include "zipmap.h"
 #include "endianconv.h"
 #include "stream.h"
+#include "storage.h"
 
 #include <math.h>
 #include <sys/types.h>
@@ -284,7 +285,7 @@ void *rdbLoadIntegerObject(rio *rdb, int enctype, int flags, size_t *lenptr) {
         char buf[LONG_STR_SIZE], *p;
         int len = ll2string(buf,sizeof(buf),val);
         if (lenptr) *lenptr = len;
-        p = plain ? zmalloc(len) : sdsnewlen(SDS_NOINIT,len);
+        p = plain ? zmalloc(len, MALLOC_SHARED) : sdsnewlen(SDS_NOINIT,len);
         memcpy(p,buf,len);
         return p;
     } else if (encode) {
@@ -345,7 +346,7 @@ ssize_t rdbSaveLzfStringObject(rio *rdb, unsigned char *s, size_t len) {
     /* We require at least four bytes compression for this to be worth it */
     if (len <= 4) return 0;
     outlen = len-4;
-    if ((out = zmalloc(outlen+1)) == NULL) return 0;
+    if ((out = zmalloc(outlen+1, MALLOC_LOCAL)) == NULL) return 0;
     comprlen = lzf_compress(s, len, out, outlen);
     if (comprlen == 0) {
         zfree(out);
@@ -368,11 +369,11 @@ void *rdbLoadLzfStringObject(rio *rdb, int flags, size_t *lenptr) {
 
     if ((clen = rdbLoadLen(rdb,NULL)) == RDB_LENERR) return NULL;
     if ((len = rdbLoadLen(rdb,NULL)) == RDB_LENERR) return NULL;
-    if ((c = zmalloc(clen)) == NULL) goto err;
+    if ((c = zmalloc(clen, MALLOC_SHARED)) == NULL) goto err;
 
     /* Allocate our target according to the uncompressed size. */
     if (plain) {
-        val = zmalloc(len);
+        val = zmalloc(len, MALLOC_SHARED);
     } else {
         val = sdsnewlen(SDS_NOINIT,len);
     }
@@ -501,7 +502,7 @@ void *rdbGenericLoadStringObject(rio *rdb, int flags, size_t *lenptr) {
 
     if (len == RDB_LENERR) return NULL;
     if (plain || sds) {
-        void *buf = plain ? zmalloc(len) : sdsnewlen(SDS_NOINIT,len);
+        void *buf = plain ? zmalloc(len, MALLOC_SHARED) : sdsnewlen(SDS_NOINIT,len);
         if (lenptr) *lenptr = len;
         if (len && rioRead(rdb,buf,len) == 0) {
             if (plain)
@@ -1290,10 +1291,12 @@ int rdbSaveBackground(char *filename, rdbSaveInfo *rsi) {
     openChildInfoPipe();
 
     start = ustime();
+    handle_prefork();
     if ((childpid = fork()) == 0) {
         int retval;
 
         /* Child */
+        handle_postfork(0);
         closeListeningSockets(0);
         redisSetProcTitle("redis-rdb-bgsave");
         retval = rdbSave(filename,rsi);
@@ -1312,6 +1315,7 @@ int rdbSaveBackground(char *filename, rdbSaveInfo *rsi) {
         exitFromChild((retval == C_OK) ? 0 : 1);
     } else {
         /* Parent */
+        handle_postfork(childpid);
         server.stat_fork_time = ustime()-start;
         server.stat_fork_rate = (double) zmalloc_used_memory() * 1000000 / server.stat_fork_time / (1024*1024*1024); /* GB per second. */
         latencyAddSampleIfNeeded("fork",server.stat_fork_time/1000);
@@ -2141,7 +2145,7 @@ void backgroundSaveDoneHandlerSocket(int exitcode, int bysignal) {
      * If the process returned an error, consider the list of slaves that
      * can continue to be empty, so that it's just a special case of the
      * normal code path. */
-    ok_slaves = zmalloc(sizeof(uint64_t)); /* Make space for the count. */
+    ok_slaves = zmalloc(sizeof(uint64_t), MALLOC_LOCAL); /* Make space for the count. */
     ok_slaves[0] = 0;
     if (!bysignal && exitcode == 0) {
         int readlen = sizeof(uint64_t);
@@ -2258,11 +2262,11 @@ int rdbSaveToSlavesSockets(rdbSaveInfo *rsi) {
 
     /* Collect the file descriptors of the slaves we want to transfer
      * the RDB to, which are i WAIT_BGSAVE_START state. */
-    fds = zmalloc(sizeof(int)*listLength(server.slaves));
+    fds = zmalloc(sizeof(int)*listLength(server.slaves), MALLOC_LOCAL);
     /* We also allocate an array of corresponding client IDs. This will
      * be useful for the child process in order to build the report
      * (sent via unix pipe) that will be sent to the parent. */
-    clientids = zmalloc(sizeof(uint64_t)*listLength(server.slaves));
+    clientids = zmalloc(sizeof(uint64_t)*listLength(server.slaves), MALLOC_LOCAL);
     numfds = 0;
 
     listRewind(server.slaves,&li);
@@ -2326,7 +2330,7 @@ int rdbSaveToSlavesSockets(rdbSaveInfo *rsi) {
              * can match the report with a specific slave, and 'error' is
              * set to 0 if the replication process terminated with a success
              * or the error code if an error occurred. */
-            void *msg = zmalloc(sizeof(uint64_t)*(1+2*numfds));
+            void *msg = zmalloc(sizeof(uint64_t)*(1+2*numfds), MALLOC_LOCAL);
             uint64_t *len = msg;
             uint64_t *ids = len+1;
             int j, msglen;
diff --git a/src/redis-benchmark.c b/src/redis-benchmark.c
index d30879dc4..716c492df 100644
--- a/src/redis-benchmark.c
+++ b/src/redis-benchmark.c
@@ -45,6 +45,7 @@
 #include "hiredis.h"
 #include "adlist.h"
 #include "zmalloc.h"
+#include "storage.h"
 
 #define UNUSED(V) ((void) V)
 #define RANDPTR_INITIAL_SIZE 8
@@ -315,7 +316,7 @@ static void writeHandler(aeEventLoop *el, int fd, void *privdata, int mask) {
  * Even when cloning another client, prefix commands are applied if needed.*/
 static client createClient(char *cmd, size_t len, client from) {
     int j;
-    client c = zmalloc(sizeof(struct _client));
+    client c = zmalloc(sizeof(struct _client), MALLOC_LOCAL);
 
     if (config.hostsocket == NULL) {
         c->context = redisConnectNonBlock(config.hostip,config.hostport);
@@ -379,7 +380,7 @@ static client createClient(char *cmd, size_t len, client from) {
         if (from) {
             c->randlen = from->randlen;
             c->randfree = 0;
-            c->randptr = zmalloc(sizeof(char*)*c->randlen);
+            c->randptr = zmalloc(sizeof(char*)*c->randlen, MALLOC_LOCAL);
             /* copy the offsets. */
             for (j = 0; j < (int)c->randlen; j++) {
                 c->randptr[j] = c->obuf + (from->randptr[j]-from->obuf);
@@ -391,7 +392,7 @@ static client createClient(char *cmd, size_t len, client from) {
 
             c->randlen = 0;
             c->randfree = RANDPTR_INITIAL_SIZE;
-            c->randptr = zmalloc(sizeof(char*)*c->randfree);
+            c->randptr = zmalloc(sizeof(char*)*c->randfree, MALLOC_LOCAL);
             while ((p = strstr(p,"__rand_int__")) != NULL) {
                 if (c->randfree == 0) {
                     c->randptr = zrealloc(c->randptr,sizeof(char*)*c->randlen*2);
@@ -652,6 +653,8 @@ int main(int argc, const char **argv) {
 
     client c;
 
+    storage_init();
+
     srandom(time(NULL));
     signal(SIGHUP, SIG_IGN);
     signal(SIGPIPE, SIG_IGN);
@@ -684,7 +687,7 @@ int main(int argc, const char **argv) {
     argc -= i;
     argv += i;
 
-    config.latency = zmalloc(sizeof(long long)*config.requests);
+    config.latency = zmalloc(sizeof(long long)*config.requests, MALLOC_LOCAL);
 
     if (config.keepalive == 0) {
         printf("WARNING: keepalive disabled, you probably need 'echo 1 > /proc/sys/net/ipv4/tcp_tw_reuse' for Linux and 'sudo sysctl -w net.inet.tcp.msl=1000' for Mac OS X in order to use a lot of clients/requests\n");
@@ -716,7 +719,7 @@ int main(int argc, const char **argv) {
     }
 
     /* Run default benchmark suite. */
-    data = zmalloc(config.datasize+1);
+    data = zmalloc(config.datasize+1, MALLOC_LOCAL);
     do {
         memset(data,'x',config.datasize);
         data[config.datasize] = '\0';
diff --git a/src/redis-check-aof.c b/src/redis-check-aof.c
index c4d5a225e..a79adb946 100644
--- a/src/redis-check-aof.c
+++ b/src/redis-check-aof.c
@@ -82,7 +82,7 @@ int readString(FILE *fp, char** target) {
 
     /* Increase length to also consume \r\n */
     len += 2;
-    *target = (char*)zmalloc(len);
+    *target = (char*)zmalloc(len, MALLOC_LOCAL);
     if (!readBytes(fp,*target,len)) {
         return 0;
     }
diff --git a/src/redis-cli.c b/src/redis-cli.c
index 93290e5ed..91deac496 100644
--- a/src/redis-cli.c
+++ b/src/redis-cli.c
@@ -55,6 +55,7 @@
 #include "help.h"
 #include "anet.h"
 #include "ae.h"
+#include "storage.h"
 
 #define UNUSED(V) ((void) V)
 
@@ -477,11 +478,11 @@ static void cliInitHelp(void) {
     helpEntry tmp;
 
     helpEntriesLen = len = commandslen+groupslen;
-    helpEntries = zmalloc(sizeof(helpEntry)*len);
+    helpEntries = zmalloc(sizeof(helpEntry)*len, MALLOC_LOCAL);
 
     for (i = 0; i < groupslen; i++) {
         tmp.argc = 1;
-        tmp.argv = zmalloc(sizeof(sds));
+        tmp.argv = zmalloc(sizeof(sds), MALLOC_LOCAL);
         tmp.argv[0] = sdscatprintf(sdsempty(),"@%s",commandGroups[i]);
         tmp.full = tmp.argv[0];
         tmp.type = CLI_HELP_GROUP;
@@ -532,13 +533,13 @@ static void cliIntegrateHelp(void) {
         helpEntry *new = helpEntries+(helpEntriesLen-1);
 
         new->argc = 1;
-        new->argv = zmalloc(sizeof(sds));
+        new->argv = zmalloc(sizeof(sds), MALLOC_LOCAL);
         new->argv[0] = sdsnew(cmdname);
         new->full = new->argv[0];
         new->type = CLI_HELP_COMMAND;
         sdstoupper(new->argv[0]);
 
-        struct commandHelp *ch = zmalloc(sizeof(*ch));
+        struct commandHelp *ch = zmalloc(sizeof(*ch), MALLOC_LOCAL);
         ch->name = new->argv[0];
         ch->params = sdsempty();
         int args = llabs(entry->element[1]->integer);
@@ -1171,7 +1172,7 @@ static int cliSendCommand(int argc, char **argv, long repeat) {
     }
 
     /* Setup argument length */
-    argvlen = zmalloc(argc*sizeof(size_t));
+    argvlen = zmalloc(argc*sizeof(size_t), MALLOC_LOCAL);
     for (j = 0; j < argc; j++)
         argvlen[j] = sdslen(argv[j]);
 
@@ -1590,7 +1591,7 @@ static int confirmWithYes(char *msg) {
 /* Turn the plain C strings into Sds strings */
 static char **convertToSds(int count, char** args) {
   int j;
-  char **sds = zmalloc(sizeof(char*)*count);
+  char **sds = zmalloc(sizeof(char*)*count, MALLOC_LOCAL);
 
   for(j = 0; j < count; j++)
     sds[j] = sdsnew(args[j]);
@@ -1872,7 +1873,7 @@ static int evalMode(int argc, char **argv) {
         }
 
         /* Create our argument vector */
-        argv2 = zmalloc(sizeof(sds)*(argc+3));
+        argv2 = zmalloc(sizeof(sds)*(argc+3), MALLOC_LOCAL);
         argv2[0] = sdsnew("EVAL");
         argv2[1] = script;
         for (j = 0; j < argc; j++) {
@@ -2193,7 +2194,7 @@ static void freeClusterManager(void) {
 }
 
 static clusterManagerNode *clusterManagerNewNode(char *ip, int port) {
-    clusterManagerNode *node = zmalloc(sizeof(*node));
+    clusterManagerNode *node = zmalloc(sizeof(*node), MALLOC_LOCAL);
     node->context = NULL;
     node->name = NULL;
     node->ip = ip;
@@ -2228,7 +2229,7 @@ static int clusterManagerCheckRedisReply(clusterManagerNode *n,
     if (!r || (is_err = (r->type == REDIS_REPLY_ERROR))) {
         if (is_err) {
             if (err != NULL) {
-                *err = zmalloc((r->len + 1) * sizeof(char));
+                *err = zmalloc((r->len + 1) * sizeof(char), MALLOC_LOCAL);
                 strcpy(*err, r->str);
             } else CLUSTER_MANAGER_PRINT_REPLY_ERROR(n, r->str);
         }
@@ -2374,7 +2375,7 @@ static redisReply *clusterManagerGetNodeRedisInfo(clusterManagerNode *node,
     if (info == NULL) return NULL;
     if (info->type == REDIS_REPLY_ERROR) {
         if (err != NULL) {
-            *err = zmalloc((info->len + 1) * sizeof(char));
+            *err = zmalloc((info->len + 1) * sizeof(char), MALLOC_LOCAL);
             strcpy(*err, info->str);
         }
         freeReplyObject(info);
@@ -2451,7 +2452,7 @@ static int clusterManagerGetAntiAffinityScore(clusterManagerNodeArray *ipnodes,
     int node_len = cluster_manager.nodes->len;
     clusterManagerNode **offending_p = NULL;
     if (offending != NULL) {
-        *offending = zcalloc(node_len * sizeof(clusterManagerNode*));
+        *offending = zcalloc(node_len * sizeof(clusterManagerNode*), MALLOC_LOCAL);
         offending_p = *offending;
     }
     /* For each set of nodes in the same host, split by
@@ -2541,7 +2542,7 @@ static void clusterManagerOptimizeAntiAffinity(clusterManagerNodeArray *ipnodes,
         clusterManagerNode *first = offenders[rand_idx],
                            *second = NULL;
         clusterManagerNode **other_replicas = zcalloc((node_len - 1) *
-                                                      sizeof(*other_replicas));
+                                                      sizeof(*other_replicas), MALLOC_LOCAL);
         int other_replicas_count = 0;
         listIter li;
         listNode *ln;
@@ -2769,8 +2770,8 @@ static int clusterManagerAddSlots(clusterManagerNode *node, char**err)
     int success = 1;
     /* First two args are used for the command itself. */
     int argc = node->slots_count + 2;
-    sds *argv = zmalloc(argc * sizeof(*argv));
-    size_t *argvlen = zmalloc(argc * sizeof(*argvlen));
+    sds *argv = zmalloc(argc * sizeof(*argv), MALLOC_LOCAL);
+    size_t *argvlen = zmalloc(argc * sizeof(*argvlen), MALLOC_LOCAL);
     argv[0] = "CLUSTER";
     argv[1] = "ADDSLOTS";
     argvlen[0] = 7;
@@ -2869,7 +2870,7 @@ static int clusterManagerSetSlot(clusterManagerNode *node1,
     if (reply->type == REDIS_REPLY_ERROR) {
         success = 0;
         if (err != NULL) {
-            *err = zmalloc((reply->len + 1) * sizeof(char));
+            *err = zmalloc((reply->len + 1) * sizeof(char), MALLOC_LOCAL);
             strcpy(*err, reply->str);
         } else CLUSTER_MANAGER_PRINT_REPLY_ERROR(node1, reply->str);
         goto cleanup;
@@ -2983,8 +2984,8 @@ static int clusterManagerCompareKeysValues(clusterManagerNode *n1,
 {
     size_t i, argc = keys_reply->elements + 2;
     static const char *hash_zero = "0000000000000000000000000000000000000000";
-    char **argv = zcalloc(argc * sizeof(char *));
-    size_t  *argv_len = zcalloc(argc * sizeof(size_t));
+    char **argv = zcalloc(argc * sizeof(char *), MALLOC_LOCAL);
+    size_t  *argv_len = zcalloc(argc * sizeof(size_t), MALLOC_LOCAL);
     argv[0] = "DEBUG";
     argv_len[0] = 5;
     argv[1] = "DIGEST-VALUE";
@@ -3051,8 +3052,8 @@ static redisReply *clusterManagerMigrateKeysInReply(clusterManagerNode *source,
     if (config.auth) c += 2;
     size_t argc = c + reply->elements;
     size_t i, offset = 6; // Keys Offset
-    argv = zcalloc(argc * sizeof(char *));
-    argv_len = zcalloc(argc * sizeof(size_t));
+    argv = zcalloc(argc * sizeof(char *), MALLOC_LOCAL);
+    argv_len = zcalloc(argc * sizeof(size_t), MALLOC_LOCAL);
     char portstr[255];
     char timeoutstr[255];
     snprintf(portstr, 10, "%d", target->port);
@@ -3130,7 +3131,7 @@ static int clusterManagerMigrateKeysInSlot(clusterManagerNode *source,
         if (reply->type == REDIS_REPLY_ERROR) {
             success = 0;
             if (err != NULL) {
-                *err = zmalloc((reply->len + 1) * sizeof(char));
+                *err = zmalloc((reply->len + 1) * sizeof(char), MALLOC_LOCAL);
                 strcpy(*err, reply->str);
                 CLUSTER_MANAGER_PRINT_REPLY_ERROR(source, *err);
             }
@@ -3142,7 +3143,7 @@ static int clusterManagerMigrateKeysInSlot(clusterManagerNode *source,
             freeReplyObject(reply);
             break;
         }
-        if (verbose) dots = zmalloc((count+1) * sizeof(char));
+        if (verbose) dots = zmalloc((count+1) * sizeof(char), MALLOC_LOCAL);
         /* Calling MIGRATE command. */
         migrate_reply = clusterManagerMigrateKeysInReply(source, target,
                                                          reply, 0, timeout,
@@ -3242,7 +3243,7 @@ static int clusterManagerMigrateKeysInSlot(clusterManagerNode *source,
             if (!success) {
                 if (migrate_reply != NULL) {
                     if (err) {
-                        *err = zmalloc((migrate_reply->len + 1) * sizeof(char));
+                        *err = zmalloc((migrate_reply->len + 1) * sizeof(char), MALLOC_LOCAL);
                         strcpy(*err, migrate_reply->str);
                     }
                     printf("\n");
@@ -3318,7 +3319,7 @@ static int clusterManagerMoveSlot(clusterManagerNode *source,
             if (r->type == REDIS_REPLY_ERROR) {
                 success = 0;
                 if (err != NULL) {
-                    *err = zmalloc((r->len + 1) * sizeof(char));
+                    *err = zmalloc((r->len + 1) * sizeof(char), MALLOC_LOCAL);
                     strcpy(*err, r->str);
                     CLUSTER_MANAGER_PRINT_REPLY_ERROR(n, *err);
                 }
@@ -3347,7 +3348,7 @@ static int clusterManagerFlushNodeConfig(clusterManagerNode *node, char **err) {
                                         node->replicate);
         if (reply == NULL || (is_err = (reply->type == REDIS_REPLY_ERROR))) {
             if (is_err && err != NULL) {
-                *err = zmalloc((reply->len + 1) * sizeof(char));
+                *err = zmalloc((reply->len + 1) * sizeof(char), MALLOC_LOCAL);
                 strcpy(*err, reply->str);
             }
             success = 0;
@@ -3715,7 +3716,7 @@ static sds clusterManagerGetConfigSignature(clusterManagerNode *node) {
                 zrealloc(node_configs, (node_count * sizeof(char *)));
             /* Make room for '|' separators. */
             tot_size += (sizeof(char) * (c - 1));
-            char *cfg = zmalloc((sizeof(char) * tot_size) + 1);
+            char *cfg = zmalloc((sizeof(char) * tot_size) + 1, MALLOC_LOCAL);
             memcpy(cfg, nodename, name_len);
             char *sp = cfg + name_len;
             *(sp++) = ':';
@@ -4641,7 +4642,7 @@ static clusterManagerNode *clusterNodeForResharding(char *id,
 static list *clusterManagerComputeReshardTable(list *sources, int numslots) {
     list *moved = listCreate();
     int src_count = listLength(sources), i = 0, tot_slots = 0, j;
-    clusterManagerNode **sorted = zmalloc(src_count * sizeof(*sorted));
+    clusterManagerNode **sorted = zmalloc(src_count * sizeof(*sorted), MALLOC_LOCAL);
     listIter li;
     listNode *ln;
     listRewind(sources, &li);
@@ -4662,7 +4663,7 @@ static list *clusterManagerComputeReshardTable(list *sources, int numslots) {
             int slot = node->slots[j];
             if (!slot) continue;
             if (count >= max || (int)listLength(moved) >= numslots) break;
-            clusterManagerReshardTableItem *item = zmalloc(sizeof(*item));
+            clusterManagerReshardTableItem *item = zmalloc(sizeof(*item), MALLOC_LOCAL);
             item->source = node;
             item->slot = j;
             listAddNodeTail(moved, item);
@@ -4720,7 +4721,7 @@ static void clusterManagerLog(int level, const char* fmt, ...) {
 static void clusterManagerNodeArrayInit(clusterManagerNodeArray *array,
                                         int alloc_len)
 {
-    array->nodes = zcalloc(alloc_len * sizeof(clusterManagerNode*));
+    array->nodes = zcalloc(alloc_len * sizeof(clusterManagerNode*), MALLOC_LOCAL);
     array->alloc = array->nodes;
     array->len = alloc_len;
     array->count = 0;
@@ -4861,9 +4862,9 @@ static int clusterManagerCommandCreate(int argc, char **argv) {
     clusterManagerLogInfo(">>> Performing hash slots allocation "
                           "on %d nodes...\n", node_len);
     int interleaved_len = 0, ip_count = 0;
-    clusterManagerNode **interleaved = zcalloc(node_len*sizeof(**interleaved));
-    char **ips = zcalloc(node_len * sizeof(char*));
-    clusterManagerNodeArray *ip_nodes = zcalloc(node_len * sizeof(*ip_nodes));
+    clusterManagerNode **interleaved = zcalloc(node_len*sizeof(**interleaved), MALLOC_LOCAL);
+    char **ips = zcalloc(node_len * sizeof(char*), MALLOC_LOCAL);
+    clusterManagerNodeArray *ip_nodes = zcalloc(node_len * sizeof(*ip_nodes), MALLOC_LOCAL);
     listIter li;
     listNode *ln;
     listRewind(cluster_manager.nodes, &li);
@@ -5500,7 +5501,7 @@ static int clusterManagerCommandRebalance(int argc, char **argv) {
         nodes_involved++;
         listAddNodeTail(involved, n);
     }
-    weightedNodes = zmalloc(nodes_involved * sizeof(clusterManagerNode *));
+    weightedNodes = zmalloc(nodes_involved * sizeof(clusterManagerNode *), MALLOC_LOCAL);
     if (weightedNodes == NULL) goto cleanup;
     /* Check cluster, only proceed if it looks sane. */
     clusterManagerCheckCluster(1);
@@ -5845,7 +5846,7 @@ static int clusterManagerCommandCall(int argc, char **argv) {
     if (!clusterManagerLoadInfoFromNode(refnode, 0)) return 0;
     argc--;
     argv++;
-    size_t *argvlen = zmalloc(argc*sizeof(size_t));
+    size_t *argvlen = zmalloc(argc*sizeof(size_t), MALLOC_LOCAL);
     clusterManagerLogInfo(">>> Calling");
     for (i = 0; i < argc; i++) {
         argvlen[i] = strlen(argv[i]);
@@ -6841,7 +6842,7 @@ static char *getInfoField(char *info, char *field) {
     n1 = strchr(p,'\r');
     n2 = strchr(p,',');
     if (n2 && n2 < n1) n1 = n2;
-    result = zmalloc(sizeof(char)*(n1-p)+1);
+    result = zmalloc(sizeof(char)*(n1-p)+1, MALLOC_LOCAL);
     memcpy(result,p,(n1-p));
     result[n1-p] = '\0';
     return result;
@@ -7175,6 +7176,7 @@ static void intrinsicLatencyMode(void) {
 int main(int argc, char **argv) {
     int firstarg;
 
+    storage_init();
     config.hostip = sdsnew("127.0.0.1");
     config.hostport = 6379;
     config.hostsocket = NULL;
diff --git a/src/replication.c b/src/replication.c
index 3bc42d62a..103af2691 100644
--- a/src/replication.c
+++ b/src/replication.c
@@ -76,7 +76,7 @@ char *replicationGetSlaveName(client *c) {
 
 void createReplicationBacklog(void) {
     serverAssert(server.repl_backlog == NULL);
-    server.repl_backlog = zmalloc(server.repl_backlog_size);
+    server.repl_backlog = zmalloc(server.repl_backlog_size, MALLOC_LOCAL);
     server.repl_backlog_histlen = 0;
     server.repl_backlog_idx = 0;
 
@@ -105,7 +105,7 @@ void resizeReplicationBacklog(long long newsize) {
          * worse often we need to alloc additional space before freeing the
          * old buffer. */
         zfree(server.repl_backlog);
-        server.repl_backlog = zmalloc(server.repl_backlog_size);
+        server.repl_backlog = zmalloc(server.repl_backlog_size, MALLOC_LOCAL);
         server.repl_backlog_histlen = 0;
         server.repl_backlog_idx = 0;
         /* Next byte we have is... the next since the buffer is empty. */
diff --git a/src/rio.c b/src/rio.c
index c9c76b8f2..0f44b017d 100644
--- a/src/rio.c
+++ b/src/rio.c
@@ -267,8 +267,8 @@ void rioInitWithFdset(rio *r, int *fds, int numfds) {
     int j;
 
     *r = rioFdsetIO;
-    r->io.fdset.fds = zmalloc(sizeof(int)*numfds);
-    r->io.fdset.state = zmalloc(sizeof(int)*numfds);
+    r->io.fdset.fds = zmalloc(sizeof(int)*numfds, MALLOC_LOCAL);
+    r->io.fdset.state = zmalloc(sizeof(int)*numfds, MALLOC_LOCAL);
     memcpy(r->io.fdset.fds,fds,sizeof(int)*numfds);
     for (j = 0; j < numfds; j++) r->io.fdset.state[j] = 0;
     r->io.fdset.numfds = numfds;
diff --git a/src/scripting.c b/src/scripting.c
index f6df38400..3b3b996e0 100644
--- a/src/scripting.c
+++ b/src/scripting.c
@@ -1866,7 +1866,7 @@ sds *ldbReplParseCommand(int *argcp) {
     if (*argcp <= 0 || *argcp > 1024) goto protoerr;
 
     /* Parse each argument. */
-    argv = zmalloc(sizeof(sds)*(*argcp));
+    argv = zmalloc(sizeof(sds)*(*argcp), MALLOC_LOCAL);
     argc = 0;
     while(argc < *argcp) {
         if (*p != '$') goto protoerr;
diff --git a/src/sds.c b/src/sds.c
index 330c955e8..a87173d94 100644
--- a/src/sds.c
+++ b/src/sds.c
@@ -96,7 +96,7 @@ sds sdsnewlen(const void *init, size_t initlen) {
     int hdrlen = sdsHdrSize(type);
     unsigned char *fp; /* flags pointer. */
 
-    sh = s_malloc(hdrlen+initlen+1);
+    sh = s_malloc(hdrlen+initlen+1, MALLOC_SHARED);
     if (init==SDS_NOINIT)
         init = NULL;
     else if (!init)
@@ -234,7 +234,7 @@ sds sdsMakeRoomFor(sds s, size_t addlen) {
     } else {
         /* Since the header size changes, need to move the string forward,
          * and can't use realloc */
-        newsh = s_malloc(hdrlen+newlen+1);
+        newsh = s_malloc(hdrlen+newlen+1, MALLOC_SHARED);
         if (newsh == NULL) return NULL;
         memcpy((char*)newsh+hdrlen, s, len+1);
         s_free(sh);
@@ -273,7 +273,7 @@ sds sdsRemoveFreeSpace(sds s) {
         if (newsh == NULL) return NULL;
         s = (char*)newsh+oldhdrlen;
     } else {
-        newsh = s_malloc(hdrlen+len+1);
+        newsh = s_malloc(hdrlen+len+1, MALLOC_SHARED);
         if (newsh == NULL) return NULL;
         memcpy((char*)newsh+hdrlen, s, len+1);
         s_free(sh);
@@ -523,7 +523,7 @@ sds sdscatvprintf(sds s, const char *fmt, va_list ap) {
     /* We try to start using a static buffer for speed.
      * If not possible we revert to heap allocation. */
     if (buflen > sizeof(staticbuf)) {
-        buf = s_malloc(buflen);
+        buf = s_malloc(buflen, MALLOC_SHARED);
         if (buf == NULL) return NULL;
     } else {
         buflen = sizeof(staticbuf);
@@ -539,7 +539,7 @@ sds sdscatvprintf(sds s, const char *fmt, va_list ap) {
         if (buf[buflen-2] != '\0') {
             if (buf != staticbuf) s_free(buf);
             buflen *= 2;
-            buf = s_malloc(buflen);
+            buf = s_malloc(buflen, MALLOC_SHARED);
             if (buf == NULL) return NULL;
             continue;
         }
@@ -816,7 +816,7 @@ sds *sdssplitlen(const char *s, ssize_t len, const char *sep, int seplen, int *c
 
     if (seplen < 1 || len < 0) return NULL;
 
-    tokens = s_malloc(sizeof(sds)*slots);
+    tokens = s_malloc(sizeof(sds)*slots, MALLOC_SHARED);
     if (tokens == NULL) return NULL;
 
     if (len == 0) {
@@ -1044,7 +1044,7 @@ sds *sdssplitargs(const char *line, int *argc) {
             current = NULL;
         } else {
             /* Even on empty input string return something not NULL. */
-            if (vector == NULL) vector = s_malloc(sizeof(void*));
+            if (vector == NULL) vector = s_malloc(sizeof(void*), MALLOC_SHARED);
             return vector;
         }
     }
@@ -1111,7 +1111,7 @@ sds sdsjoinsds(sds *argv, int argc, const char *sep, size_t seplen) {
  * the overhead of function calls. Here we define these wrappers only for
  * the programs SDS is linked to, if they want to touch the SDS internals
  * even if they use a different allocator. */
-void *sds_malloc(size_t size) { return s_malloc(size); }
+void *sds_malloc(size_t size) { return s_malloc(size, MALLOC_SHARED); }
 void *sds_realloc(void *ptr, size_t size) { return s_realloc(ptr,size); }
 void sds_free(void *ptr) { s_free(ptr); }
 
diff --git a/src/sdsalloc.h b/src/sdsalloc.h
index 531d41929..cb814acce 100644
--- a/src/sdsalloc.h
+++ b/src/sdsalloc.h
@@ -37,6 +37,10 @@
  * to use the default libc allocator). */
 
 #include "zmalloc.h"
-#define s_malloc zmalloc
-#define s_realloc zrealloc
-#define s_free zfree
+#include "storage.h"
+//#define s_malloc zmalloc
+#define s_malloc salloc
+//#define s_realloc zrealloc
+#define s_realloc srealloc
+//#define s_free zfree
+#define s_free sfree
diff --git a/src/sentinel.c b/src/sentinel.c
index 4d03c9c12..7add1e744 100644
--- a/src/sentinel.c
+++ b/src/sentinel.c
@@ -341,7 +341,7 @@ static int redisAeAttach(aeEventLoop *loop, redisAsyncContext *ac) {
         return C_ERR;
 
     /* Create container for context and r/w events */
-    e = (redisAeEvents*)zmalloc(sizeof(*e));
+    e = (redisAeEvents*)zmalloc(sizeof(*e), MALLOC_LOCAL);
     e->context = ac;
     e->loop = loop;
     e->fd = c->fd;
@@ -549,7 +549,7 @@ sentinelAddr *createSentinelAddr(char *hostname, int port) {
         errno = ENOENT;
         return NULL;
     }
-    sa = zmalloc(sizeof(*sa));
+    sa = zmalloc(sizeof(*sa), MALLOC_LOCAL);
     sa->ip = sdsnew(ip);
     sa->port = port;
     return sa;
@@ -559,7 +559,7 @@ sentinelAddr *createSentinelAddr(char *hostname, int port) {
 sentinelAddr *dupSentinelAddr(sentinelAddr *src) {
     sentinelAddr *sa;
 
-    sa = zmalloc(sizeof(*sa));
+    sa = zmalloc(sizeof(*sa), MALLOC_LOCAL);
     sa->ip = sdsnew(src->ip);
     sa->port = src->port;
     return sa;
@@ -703,10 +703,10 @@ void sentinelScheduleScriptExecution(char *path, ...) {
     va_end(ap);
     argv[0] = sdsnew(path);
 
-    sj = zmalloc(sizeof(*sj));
+    sj = zmalloc(sizeof(*sj), MALLOC_LOCAL);
     sj->flags = SENTINEL_SCRIPT_NONE;
     sj->retry_num = 0;
-    sj->argv = zmalloc(sizeof(char*)*(argc+1));
+    sj->argv = zmalloc(sizeof(char*)*(argc+1), MALLOC_LOCAL);
     sj->start_time = 0;
     sj->pid = 0;
     memcpy(sj->argv,argv,sizeof(char*)*(argc+1));
@@ -950,7 +950,7 @@ void sentinelCallClientReconfScript(sentinelRedisInstance *master, int role, cha
 
 /* Create a not yet connected link object. */
 instanceLink *createInstanceLink(void) {
-    instanceLink *link = zmalloc(sizeof(*link));
+    instanceLink *link = zmalloc(sizeof(*link), MALLOC_LOCAL);
 
     link->refcount = 1;
     link->disconnected = 1;
@@ -1200,7 +1200,7 @@ sentinelRedisInstance *createSentinelRedisInstance(char *name, int flags, char *
     }
 
     /* Create the instance object. */
-    ri = zmalloc(sizeof(*ri));
+    ri = zmalloc(sizeof(*ri), MALLOC_LOCAL);
     /* Note that all the instances are started in the disconnected state,
      * the event loop will take care of connecting them. */
     ri->flags = flags;
@@ -4091,7 +4091,7 @@ int compareSlavesForPromotion(const void *a, const void *b) {
 
 sentinelRedisInstance *sentinelSelectSlave(sentinelRedisInstance *master) {
     sentinelRedisInstance **instance =
-        zmalloc(sizeof(instance[0])*dictSize(master->slaves));
+        zmalloc(sizeof(instance[0])*dictSize(master->slaves), MALLOC_LOCAL);
     sentinelRedisInstance *selected = NULL;
     int instances = 0;
     dictIterator *di;
diff --git a/src/server.c b/src/server.c
index 1df63f65f..62b37dfa1 100644
--- a/src/server.c
+++ b/src/server.c
@@ -33,6 +33,7 @@
 #include "bio.h"
 #include "latency.h"
 #include "atomicvar.h"
+#include "storage.h"
 
 #include <time.h>
 #include <signal.h>
@@ -2729,7 +2730,7 @@ void initServer(void) {
             strerror(errno));
         exit(1);
     }
-    server.db = zmalloc(sizeof(redisDb)*server.dbnum);
+    server.db = zmalloc(sizeof(redisDb)*server.dbnum, MALLOC_LOCAL);
 
     /* Open the TCP listening socket for the user commands. */
     if (server.port != 0 &&
@@ -3066,7 +3067,7 @@ void alsoPropagate(struct redisCommand *cmd, int dbid, robj **argv, int argc,
 
     if (server.loading) return; /* No propagation during loading. */
 
-    argvcopy = zmalloc(sizeof(robj*)*argc);
+    argvcopy = zmalloc(sizeof(robj*)*argc, MALLOC_LOCAL);
     for (j = 0; j < argc; j++) {
         argvcopy[j] = argv[j];
         incrRefCount(argv[j]);
@@ -4481,7 +4482,7 @@ void usage(void) {
 
 void redisAsciiArt(void) {
 #include "asciilogo.h"
-    char *buf = zmalloc(1024*16);
+    char *buf = zmalloc(1024*16, MALLOC_LOCAL);
     char *mode;
 
     if (server.cluster_enabled) mode = "cluster";
@@ -4739,6 +4740,8 @@ int main(int argc, char **argv) {
     struct timeval tv;
     int j;
 
+    storage_init();
+
 #ifdef REDIS_TEST
     if (argc == 3 && !strcasecmp(argv[1], "test")) {
         if (!strcasecmp(argv[2], "ziplist")) {
@@ -4789,7 +4792,7 @@ int main(int argc, char **argv) {
     /* Store the executable path and arguments in a safe place in order
      * to be able to restart the server later. */
     server.executable = getAbsolutePath(argv[0]);
-    server.exec_argv = zmalloc(sizeof(char*)*(argc+1));
+    server.exec_argv = zmalloc(sizeof(char*)*(argc+1), MALLOC_LOCAL);
     server.exec_argv[argc] = NULL;
     for (j = 0; j < argc; j++) server.exec_argv[j] = zstrdup(argv[j]);
 
diff --git a/src/slowlog.c b/src/slowlog.c
index 1d715e39b..b6041f096 100644
--- a/src/slowlog.c
+++ b/src/slowlog.c
@@ -46,12 +46,12 @@
  * Incrementing the ref count of all the objects retained is up to
  * this function. */
 slowlogEntry *slowlogCreateEntry(client *c, robj **argv, int argc, long long duration) {
-    slowlogEntry *se = zmalloc(sizeof(*se));
+    slowlogEntry *se = zmalloc(sizeof(*se), MALLOC_LOCAL);
     int j, slargc = argc;
 
     if (slargc > SLOWLOG_ENTRY_MAX_ARGC) slargc = SLOWLOG_ENTRY_MAX_ARGC;
     se->argc = slargc;
-    se->argv = zmalloc(sizeof(robj*)*slargc);
+    se->argv = zmalloc(sizeof(robj*)*slargc, MALLOC_LOCAL);
     for (j = 0; j < slargc; j++) {
         /* Logging too many arguments is a useless memory waste, so we stop
          * at SLOWLOG_ENTRY_MAX_ARGC, but use the last argument to specify
diff --git a/src/sort.c b/src/sort.c
index 8608cd8b3..7ae794501 100644
--- a/src/sort.c
+++ b/src/sort.c
@@ -36,7 +36,7 @@
 zskiplistNode* zslGetElementByRank(zskiplist *zsl, unsigned long rank);
 
 redisSortOperation *createSortOperation(int type, robj *pattern) {
-    redisSortOperation *so = zmalloc(sizeof(*so));
+    redisSortOperation *so = zmalloc(sizeof(*so), MALLOC_LOCAL);
     so->type = type;
     so->pattern = pattern;
     return so;
@@ -340,7 +340,7 @@ void sortCommand(client *c) {
     }
 
     /* Load the sorting vector with all the objects to sort */
-    vector = zmalloc(sizeof(redisSortObject)*vectorlen);
+    vector = zmalloc(sizeof(redisSortObject)*vectorlen, MALLOC_LOCAL);
     j = 0;
 
     if (sortval->type == OBJ_LIST && dontsort) {
diff --git a/src/sparkline.c b/src/sparkline.c
index 0a986883d..e18ce98fd 100644
--- a/src/sparkline.c
+++ b/src/sparkline.c
@@ -55,7 +55,7 @@ static int label_margin_top = 1;
 
 /* Create a new sequence. */
 struct sequence *createSparklineSequence(void) {
-    struct sequence *seq = zmalloc(sizeof(*seq));
+    struct sequence *seq = zmalloc(sizeof(*seq), MALLOC_LOCAL);
     seq->length = 0;
     seq->samples = NULL;
     return seq;
@@ -99,7 +99,7 @@ sds sparklineRenderRange(sds output, struct sequence *seq, int rows, int offset,
     double relmax = seq->max - seq->min;
     int steps = charset_len*rows;
     int row = 0;
-    char *chars = zmalloc(len);
+    char *chars = zmalloc(len, MALLOC_LOCAL);
     int loop = 1;
     int opt_fill = flags & SPARKLINE_FILL;
     int opt_log = flags & SPARKLINE_LOG_SCALE;
diff --git a/src/storage.c b/src/storage.c
new file mode 100644
index 000000000..90009c2a8
--- /dev/null
+++ b/src/storage.c
@@ -0,0 +1,155 @@
+#include "server.h"
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <memkind.h>
+#include <sys/ioctl.h>
+#include <linux/fs.h>
+#include <unistd.h>
+#include <inttypes.h>
+#include "storage.h"
+
+struct memkind *mkdisk = NULL;
+static char *PMEM_DIR = "/mnt/btrfs_scratch/";
+
+
+
+#define OBJECT_PAGE_BUFFER_SIZE 8192 //(size in objs)
+struct object_page
+{
+    uint64_t allocmap[OBJECT_PAGE_BUFFER_SIZE/(8*sizeof(uint64_t))];
+    robj rgobj[OBJECT_PAGE_BUFFER_SIZE];
+    struct object_page *pnext;
+};
+#define OBJ_PAGE_BITS_PER_WORD 64
+struct object_page *headObjpage = NULL;
+
+void storage_init()
+{
+    int errv = memkind_create_pmem(PMEM_DIR, 0, &mkdisk);
+    if (errv)
+    {
+        fprintf(stderr, "Memory pool creation failed: %d\n", errv);
+        exit(EXIT_FAILURE);
+    }
+    headObjpage = memkind_calloc(MEMKIND_HUGETLB, 1, sizeof(struct object_page));
+}
+
+int IdxAllocObject(struct object_page *page)
+{
+    for (size_t iword = 0; iword < OBJ_PAGE_BITS_PER_WORD; ++iword)
+    {
+        if ((page->allocmap[iword] + 1) != 0)
+        {
+            int ibit = 0;
+            uint64_t bitword = page->allocmap[iword];
+            while (bitword & 1)
+            {
+                bitword >>= 1;
+                ++ibit;
+            }
+            page->allocmap[iword] |= 1 << ibit;
+            return (iword * OBJ_PAGE_BITS_PER_WORD) + ibit;
+        }
+    }
+    return -1;
+}
+
+struct redisObject *salloc_obj()
+{
+    struct object_page *cur = headObjpage;
+    for (;;)
+    {
+        int idx = IdxAllocObject(cur);
+        if (idx >= 0)
+        {
+            return &cur->rgobj[idx];
+        }
+
+        if (cur->pnext == NULL)
+        {
+            cur->pnext = memkind_calloc(MEMKIND_HUGETLB, 1, sizeof(struct object_page));
+        }
+
+        cur = cur->pnext;
+    }
+}
+void sfree_obj(struct redisObject *obj)
+{
+    struct object_page *cur = headObjpage;
+    for (;;)
+    {
+        if (obj >= cur->rgobj && (obj < (cur->rgobj + OBJECT_PAGE_BUFFER_SIZE)))
+        {
+            // Its on this page
+            int idx = obj - cur->rgobj;
+            cur->allocmap[idx / OBJ_PAGE_BITS_PER_WORD] &= ~(1 << (idx % OBJ_PAGE_BITS_PER_WORD));
+            break;
+        }
+        cur = cur->pnext;
+    }
+    return;
+}
+
+void *salloc(size_t cb, enum MALLOC_CLASS class)
+{
+    switch (class)
+    {
+    case MALLOC_SHARED:
+        return memkind_malloc(mkdisk, cb);
+    default:
+        return memkind_malloc(MEMKIND_DEFAULT, cb);
+    }
+    return NULL;
+}
+
+void *scalloc(size_t cb, size_t c, enum MALLOC_CLASS class)
+{
+    switch (class)
+    {
+    case MALLOC_SHARED:
+        return memkind_calloc(mkdisk, cb, c);
+    default:
+        return memkind_calloc(MEMKIND_DEFAULT, cb, c);
+    }
+    return NULL;
+}
+
+void sfree(void *pv)
+{
+    memkind_free(NULL, pv);
+}
+
+void *srealloc(void *pv, size_t cb)
+{
+    memkind_t kind = mkdisk;
+    return memkind_realloc(kind, pv, cb);
+}
+
+int fdNew = -1;
+void handle_prefork()
+{
+    memkind_tmpfile(PMEM_DIR, &fdNew);
+    if (ioctl(fdNew, FICLONE, memkind_fd(mkdisk)) == -1)
+    {
+        perror("failed to fork file");
+        exit(EXIT_FAILURE);
+    }
+}
+
+void handle_postfork(int pid)
+{
+    if (pid != 0)
+    {
+        // Parent, close fdNew
+        close(fdNew);
+        fdNew = -1;
+    }
+    else
+    {
+        int fdOriginal = memkind_fd(mkdisk);
+        memkind_pmem_remapfd(mkdisk, fdNew);
+        close(fdOriginal);
+    }
+    
+}
\ No newline at end of file
diff --git a/src/storage.h b/src/storage.h
new file mode 100644
index 000000000..ebb2d1c5e
--- /dev/null
+++ b/src/storage.h
@@ -0,0 +1,23 @@
+#ifndef __STORAGE_H__
+#define __STORAGE_H__
+
+enum MALLOC_CLASS
+{
+    MALLOC_LOCAL,
+    MALLOC_SHARED,
+};
+
+void storage_init(void);
+
+struct redisObject *salloc_obj();
+void sfree_obj(struct redisObject *obj);
+
+void *salloc(size_t cb, enum MALLOC_CLASS class);
+void *scalloc(size_t cb, size_t c, enum MALLOC_CLASS class);
+void sfree(void*);
+void *srealloc(void *pv, size_t cb);
+
+void handle_prefork();
+void handle_postfork(int pid);
+
+#endif
diff --git a/src/t_hash.c b/src/t_hash.c
index d8aee6572..c94945a1f 100644
--- a/src/t_hash.c
+++ b/src/t_hash.c
@@ -322,7 +322,7 @@ unsigned long hashTypeLength(const robj *o) {
 }
 
 hashTypeIterator *hashTypeInitIterator(robj *subject) {
-    hashTypeIterator *hi = zmalloc(sizeof(hashTypeIterator));
+    hashTypeIterator *hi = zmalloc(sizeof(hashTypeIterator), MALLOC_LOCAL);
     hi->subject = subject;
     hi->encoding = subject->encoding;
 
diff --git a/src/t_list.c b/src/t_list.c
index 451ffb4b5..143f672a4 100644
--- a/src/t_list.c
+++ b/src/t_list.c
@@ -82,7 +82,7 @@ unsigned long listTypeLength(const robj *subject) {
 /* Initialize an iterator at the specified index. */
 listTypeIterator *listTypeInitIterator(robj *subject, long index,
                                        unsigned char direction) {
-    listTypeIterator *li = zmalloc(sizeof(listTypeIterator));
+    listTypeIterator *li = zmalloc(sizeof(listTypeIterator), MALLOC_LOCAL);
     li->subject = subject;
     li->encoding = subject->encoding;
     li->direction = direction;
diff --git a/src/t_set.c b/src/t_set.c
index 61013dbcd..89b5cf70a 100644
--- a/src/t_set.c
+++ b/src/t_set.c
@@ -119,7 +119,7 @@ int setTypeIsMember(robj *subject, sds value) {
 }
 
 setTypeIterator *setTypeInitIterator(robj *subject) {
-    setTypeIterator *si = zmalloc(sizeof(setTypeIterator));
+    setTypeIterator *si = zmalloc(sizeof(setTypeIterator), MALLOC_LOCAL);
     si->subject = subject;
     si->encoding = subject->encoding;
     if (si->encoding == OBJ_ENCODING_HT) {
@@ -791,7 +791,7 @@ int qsortCompareSetsByRevCardinality(const void *s1, const void *s2) {
 
 void sinterGenericCommand(client *c, robj **setkeys,
                           unsigned long setnum, robj *dstkey) {
-    robj **sets = zmalloc(sizeof(robj*)*setnum);
+    robj **sets = zmalloc(sizeof(robj*)*setnum, MALLOC_SHARED);
     setTypeIterator *si;
     robj *dstset = NULL;
     sds elesds;
@@ -930,7 +930,7 @@ void sinterstoreCommand(client *c) {
 
 void sunionDiffGenericCommand(client *c, robj **setkeys, int setnum,
                               robj *dstkey, int op) {
-    robj **sets = zmalloc(sizeof(robj*)*setnum);
+    robj **sets = zmalloc(sizeof(robj*)*setnum, MALLOC_SHARED);
     setTypeIterator *si;
     robj *dstset = NULL;
     sds ele;
diff --git a/src/t_stream.c b/src/t_stream.c
index 1a5acac42..5fb14628b 100644
--- a/src/t_stream.c
+++ b/src/t_stream.c
@@ -50,7 +50,7 @@ size_t streamReplyWithRangeFromConsumerPEL(client *c, stream *s, streamID *start
 
 /* Create a new stream data structure. */
 stream *streamNew(void) {
-    stream *s = zmalloc(sizeof(*s));
+    stream *s = zmalloc(sizeof(*s), MALLOC_SHARED);
     s->rax = raxNew();
     s->length = 0;
     s->last_id.ms = 0;
@@ -1402,8 +1402,8 @@ void xreadCommand(client *c) {
 
     /* Parse the IDs and resolve the group name. */
     if (streams_count > STREAMID_STATIC_VECTOR_LEN)
-        ids = zmalloc(sizeof(streamID)*streams_count);
-    if (groupname) groups = zmalloc(sizeof(streamCG*)*streams_count);
+        ids = zmalloc(sizeof(streamID)*streams_count, MALLOC_SHARED);
+    if (groupname) groups = zmalloc(sizeof(streamCG*)*streams_count, MALLOC_SHARED);
 
     for (int i = streams_arg + streams_count; i < c->argc; i++) {
         /* Specifying "$" as last-known-id means that the client wants to be
@@ -1595,7 +1595,7 @@ cleanup: /* Cleanup. */
  * time to the current time. The NACK consumer will be set to the one
  * specified as argument of the function. */
 streamNACK *streamCreateNACK(streamConsumer *consumer) {
-    streamNACK *nack = zmalloc(sizeof(*nack));
+    streamNACK *nack = zmalloc(sizeof(*nack), MALLOC_SHARED);
     nack->delivery_time = mstime();
     nack->delivery_count = 1;
     nack->consumer = consumer;
@@ -1628,7 +1628,7 @@ streamCG *streamCreateCG(stream *s, char *name, size_t namelen, streamID *id) {
     if (raxFind(s->cgroups,(unsigned char*)name,namelen) != raxNotFound)
         return NULL;
 
-    streamCG *cg = zmalloc(sizeof(*cg));
+    streamCG *cg = zmalloc(sizeof(*cg), MALLOC_SHARED);
     cg->pel = raxNew();
     cg->consumers = raxNew();
     cg->last_id = *id;
@@ -1661,7 +1661,7 @@ streamConsumer *streamLookupConsumer(streamCG *cg, sds name, int create) {
                                sdslen(name));
     if (consumer == raxNotFound) {
         if (!create) return NULL;
-        consumer = zmalloc(sizeof(*consumer));
+        consumer = zmalloc(sizeof(*consumer), MALLOC_SHARED);
         consumer->name = sdsdup(name);
         consumer->pel = raxNew();
         raxInsert(cg->consumers,(unsigned char*)name,sdslen(name),
diff --git a/src/t_zset.c b/src/t_zset.c
index 0427ee887..daeabcf93 100644
--- a/src/t_zset.c
+++ b/src/t_zset.c
@@ -70,7 +70,7 @@ int zslLexValueLteMax(sds value, zlexrangespec *spec);
  * The SDS string 'ele' is referenced by the node after the call. */
 zskiplistNode *zslCreateNode(int level, double score, sds ele) {
     zskiplistNode *zn =
-        zmalloc(sizeof(*zn)+level*sizeof(struct zskiplistLevel));
+        zmalloc(sizeof(*zn)+level*sizeof(struct zskiplistLevel), MALLOC_SHARED);
     zn->score = score;
     zn->ele = ele;
     return zn;
@@ -81,7 +81,7 @@ zskiplist *zslCreate(void) {
     int j;
     zskiplist *zsl;
 
-    zsl = zmalloc(sizeof(*zsl));
+    zsl = zmalloc(sizeof(*zsl), MALLOC_SHARED);
     zsl->level = 1;
     zsl->length = 0;
     zsl->header = zslCreateNode(ZSKIPLIST_MAXLEVEL,0,NULL);
@@ -1182,7 +1182,7 @@ void zsetConvert(robj *zobj, int encoding) {
         if (encoding != OBJ_ENCODING_SKIPLIST)
             serverPanic("Unknown target encoding");
 
-        zs = zmalloc(sizeof(*zs));
+        zs = zmalloc(sizeof(*zs), MALLOC_SHARED);
         zs->dict = dictCreate(&zsetDictType,NULL);
         zs->zsl = zslCreate();
 
@@ -1590,7 +1590,7 @@ void zaddGenericCommand(client *c, int flags) {
     /* Start parsing all the scores, we need to emit any syntax error
      * before executing additions to the sorted set, as the command should
      * either execute fully or nothing at all. */
-    scores = zmalloc(sizeof(double)*elements);
+    scores = zmalloc(sizeof(double)*elements, MALLOC_SHARED);
     for (j = 0; j < elements; j++) {
         if (getDoubleFromObjectOrReply(c,c->argv[scoreidx+j*2],&scores[j],NULL)
             != C_OK) goto cleanup;
@@ -2203,7 +2203,7 @@ void zunionInterGenericCommand(client *c, robj *dstkey, int op) {
     }
 
     /* read keys to be used for input */
-    src = zcalloc(sizeof(zsetopsrc) * setnum);
+    src = zcalloc(sizeof(zsetopsrc) * setnum, MALLOC_SHARED);
     for (i = 0, j = 3; i < setnum; i++, j++) {
         robj *obj = lookupKeyWrite(c->db,c->argv[j]);
         if (obj != NULL) {
diff --git a/src/ziplist.c b/src/ziplist.c
index 1579d1109..733e2e685 100644
--- a/src/ziplist.c
+++ b/src/ziplist.c
@@ -577,7 +577,7 @@ void zipEntry(unsigned char *p, zlentry *e) {
 /* Create a new empty ziplist. */
 unsigned char *ziplistNew(void) {
     unsigned int bytes = ZIPLIST_HEADER_SIZE+1;
-    unsigned char *zl = zmalloc(bytes);
+    unsigned char *zl = zmalloc(bytes, MALLOC_SHARED);
     ZIPLIST_BYTES(zl) = intrev32ifbe(bytes);
     ZIPLIST_TAIL_OFFSET(zl) = intrev32ifbe(ZIPLIST_HEADER_SIZE);
     ZIPLIST_LENGTH(zl) = 0;
diff --git a/src/zipmap.c b/src/zipmap.c
index 22bfa1a46..88e7f8706 100644
--- a/src/zipmap.c
+++ b/src/zipmap.c
@@ -94,7 +94,7 @@
 
 /* Create a new empty zipmap. */
 unsigned char *zipmapNew(void) {
-    unsigned char *zm = zmalloc(2);
+    unsigned char *zm = zmalloc(2, MALLOC_SHARED);
 
     zm[0] = 0; /* Length */
     zm[1] = ZIPMAP_END;
diff --git a/src/zmalloc.c b/src/zmalloc.c
index 2482f512b..7d307bad7 100644
--- a/src/zmalloc.c
+++ b/src/zmalloc.c
@@ -70,6 +70,16 @@ void zlibc_free(void *ptr) {
 #define mallocx(size,flags) je_mallocx(size,flags)
 #define dallocx(ptr,flags) je_dallocx(ptr,flags)
 #endif
+#include "storage.h"
+#undef malloc
+#undef calloc
+#undef realloc
+#undef free
+#define malloc(size, type) salloc(size, type)
+#define calloc(count, size, type) scalloc(count, size, type)
+#define realloc(ptr, size) srealloc(ptr, size)
+#define free(ptr) sfree(ptr)
+//#define zmalloc_size(ptr) (sizeof(ptr))
 
 #define update_zmalloc_stat_alloc(__n) do { \
     size_t _n = (__n); \
@@ -95,8 +105,8 @@ static void zmalloc_default_oom(size_t size) {
 
 static void (*zmalloc_oom_handler)(size_t) = zmalloc_default_oom;
 
-void *zmalloc(size_t size) {
-    void *ptr = malloc(size+PREFIX_SIZE);
+void *zmalloc(size_t size, enum MALLOC_CLASS class) {
+    void *ptr = malloc(size+PREFIX_SIZE, class);
 
     if (!ptr) zmalloc_oom_handler(size);
 #ifdef HAVE_MALLOC_SIZE
@@ -127,8 +137,8 @@ void zfree_no_tcache(void *ptr) {
 }
 #endif
 
-void *zcalloc(size_t size) {
-    void *ptr = calloc(1, size+PREFIX_SIZE);
+void *zcalloc(size_t size, enum MALLOC_CLASS class) {
+    void *ptr = calloc(1, size+PREFIX_SIZE, class);
 
     if (!ptr) zmalloc_oom_handler(size);
 #ifdef HAVE_MALLOC_SIZE
@@ -148,7 +158,7 @@ void *zrealloc(void *ptr, size_t size) {
     size_t oldsize;
     void *newptr;
 
-    if (ptr == NULL) return zmalloc(size);
+    if (ptr == NULL) return zmalloc(size, MALLOC_SHARED);
 #ifdef HAVE_MALLOC_SIZE
     oldsize = zmalloc_size(ptr);
     newptr = realloc(ptr,size);
@@ -207,7 +217,7 @@ void zfree(void *ptr) {
 
 char *zstrdup(const char *s) {
     size_t l = strlen(s)+1;
-    char *p = zmalloc(l);
+    char *p = zmalloc(l, MALLOC_SHARED);
 
     memcpy(p,s,l);
     return p;
diff --git a/src/zmalloc.h b/src/zmalloc.h
index 6fb19b046..0692e64c2 100644
--- a/src/zmalloc.h
+++ b/src/zmalloc.h
@@ -35,7 +35,12 @@
 #define __xstr(s) __str(s)
 #define __str(s) #s
 
-#if defined(USE_TCMALLOC)
+#if 1
+    #define ZMALLOC_LIB ("custom")
+    #include "storage.h"
+    #undef USE_JEMALLOC
+    #define USE_MALLOC_CLASS 1
+#elif defined(USE_TCMALLOC)
 #define ZMALLOC_LIB ("tcmalloc-" __xstr(TC_VERSION_MAJOR) "." __xstr(TC_VERSION_MINOR))
 #include <google/tcmalloc.h>
 #if (TC_VERSION_MAJOR == 1 && TC_VERSION_MINOR >= 6) || (TC_VERSION_MAJOR > 1)
@@ -77,8 +82,8 @@
 #define HAVE_DEFRAG
 #endif
 
-void *zmalloc(size_t size);
-void *zcalloc(size_t size);
+void *zmalloc(size_t size, enum MALLOC_CLASS class);
+void *zcalloc(size_t size, enum MALLOC_CLASS class);
 void *zrealloc(void *ptr, size_t size);
 void zfree(void *ptr);
 char *zstrdup(const char *s);
diff --git a/utils/hashtable/rehashing.c b/utils/hashtable/rehashing.c
index b57a9043a..ee4d10c2d 100644
--- a/utils/hashtable/rehashing.c
+++ b/utils/hashtable/rehashing.c
@@ -71,7 +71,7 @@ int sortPointers(const void *a, const void *b) {
 void stressGetKeys(dict *d, int times, int *perfect_run, int *approx_run) {
     int j;
 
-    dictEntry **des = zmalloc(sizeof(dictEntry*)*dictSize(d));
+    dictEntry **des = zmalloc(sizeof(dictEntry*)*dictSize(d), MALLOC_SHARED);
     for (j = 0; j < times; j++) {
         int requested = rand() % (dictSize(d)+1);
         int returned = dictGetSomeKeys(d, des, requested);