diff --git a/.gitignore b/.gitignore
index f4b816d05..de4716a66 100644
--- a/.gitignore
+++ b/.gitignore
@@ -70,6 +70,7 @@ architecture.txt
 deps/jemalloc/jemalloc-3.6.0/
 deps/jemalloc/jemalloc-4.1.0/
 deps/jemalloc/jemalloc-4.2.1/
+deps/jemalloc/jemalloc-4.3.1/
 
 #libdaemon
 deps/libdaemon/libdaemon-0.14/
diff --git a/include/thread.h b/include/thread.h
index 874591767..a428e1f8c 100644
--- a/include/thread.h
+++ b/include/thread.h
@@ -24,6 +24,7 @@
 #define __thread_h__
 
 #include <pthread.h>
+#include "jemalloc.h"
 
 class Thread
 {
@@ -31,7 +32,7 @@ class Thread
     Thread();
     virtual ~Thread();
 
-    int start();
+    int start(bool jemalloc_tcache=true);
     int join();
     int detach();
     pthread_t self();
diff --git a/lib/MySQL_Monitor.cpp b/lib/MySQL_Monitor.cpp
index 5fa6ba543..de2161078 100644
--- a/lib/MySQL_Monitor.cpp
+++ b/lib/MySQL_Monitor.cpp
@@ -262,6 +262,34 @@ MySQL_Monitor_State_Data::~MySQL_Monitor_State_Data() {
 	}
 }
 
+void * monitor_connect_pthread(void *arg) {
+	bool cache=false;
+	mallctl("thread.tcache.enabled", NULL, NULL, &cache, sizeof(bool));
+	GloMyMon->monitor_connect();
+	return NULL;
+}
+
+void * monitor_ping_pthread(void *arg) {
+	bool cache=false;
+	mallctl("thread.tcache.enabled", NULL, NULL, &cache, sizeof(bool));
+	GloMyMon->monitor_ping();
+	return NULL;
+}
+
+void * monitor_read_only_pthread(void *arg) {
+	bool cache=false;
+//	mallctl("thread.tcache.enabled", NULL, NULL, &cache, sizeof(bool));
+	GloMyMon->monitor_read_only();
+	return NULL;
+}
+
+void * monitor_replication_lag_pthread(void *arg) {
+	bool cache=false;
+//	mallctl("thread.tcache.enabled", NULL, NULL, &cache, sizeof(bool));
+	GloMyMon->monitor_replication_lag();
+	return NULL;
+}
+
 MySQL_Monitor::MySQL_Monitor() {
 
 	GloMyMon = this;
@@ -1448,12 +1476,19 @@ __monitor_run:
 	ConsumerThread **threads= (ConsumerThread **)malloc(sizeof(ConsumerThread *)*num_threads);
 	for (unsigned int i=0;i<num_threads; i++) {
 		threads[i] = new ConsumerThread(queue, 0);
-		threads[i]->start();
+		threads[i]->start(false);
 	}
-	std::thread * monitor_connect_thread = new std::thread(&MySQL_Monitor::monitor_connect,this);
-	std::thread * monitor_ping_thread = new std::thread(&MySQL_Monitor::monitor_ping,this);
-	std::thread * monitor_read_only_thread = new std::thread(&MySQL_Monitor::monitor_read_only,this);
-	std::thread * monitor_replication_lag_thread = new std::thread(&MySQL_Monitor::monitor_replication_lag,this);
+	pthread_attr_t attr;
+	pthread_attr_init(&attr);
+	pthread_attr_setstacksize (&attr, 64*1024);
+	pthread_t monitor_connect_thread;
+	pthread_create(&monitor_connect_thread, &attr, &monitor_connect_pthread,NULL);
+	pthread_t monitor_ping_thread;
+	pthread_create(&monitor_ping_thread, &attr, &monitor_ping_pthread,NULL);
+	pthread_t monitor_read_only_thread;
+	pthread_create(&monitor_read_only_thread, &attr, &monitor_read_only_pthread,NULL);
+	pthread_t monitor_replication_lag_thread;
+	pthread_create(&monitor_replication_lag_thread, &attr, &monitor_replication_lag_pthread,NULL);
 	while (shutdown==false && mysql_thread___monitor_enabled==true) {
 		unsigned int glover;
 		if (GloMTH) {
@@ -1492,10 +1527,10 @@ __monitor_run:
 		threads[i]->join();
 	}
 	free(threads);
-	monitor_connect_thread->join();
-	monitor_ping_thread->join();
-	monitor_read_only_thread->join();
-	monitor_replication_lag_thread->join();
+	pthread_join(monitor_connect_thread,NULL);
+	pthread_join(monitor_ping_thread,NULL);
+	pthread_join(monitor_read_only_thread,NULL);
+	pthread_join(monitor_replication_lag_thread,NULL);
 	while (shutdown==false) {
 		unsigned int glover;
 		if (GloMTH) {
diff --git a/lib/thread.cpp b/lib/thread.cpp
index fb3403d77..32229a201 100644
--- a/lib/thread.cpp
+++ b/lib/thread.cpp
@@ -39,12 +39,19 @@ Thread::~Thread()
     }
 }
 
-int Thread::start()
+int Thread::start(bool jemalloc_tcache)
 {
-    int result = pthread_create(&m_tid, NULL, runThread, this);
+    pthread_attr_t attr;
+    pthread_attr_init(&attr);
+    pthread_attr_setstacksize (&attr, 64*1024);
+    int result = pthread_create(&m_tid, &attr, runThread, this);
     if (result == 0) {
         m_running = 1;
     }
+		if (jemalloc_tcache==false) {
+			bool cache=false;
+			mallctl("thread.tcache.enabled", NULL, NULL, &cache, sizeof(bool));
+		}
     return result;
 }
 
diff --git a/src/main.cpp b/src/main.cpp
index 17409792e..b8a888a0f 100644
--- a/src/main.cpp
+++ b/src/main.cpp
@@ -129,7 +129,8 @@ static volatile int load_;
 
 //__thread l_sfp *__thr_sfp=NULL;
 
-const char *malloc_conf = "xmalloc:true,lg_tcache_max:16,purge:decay";
+//const char *malloc_conf = "xmalloc:true,lg_tcache_max:16,purge:decay";
+const char *malloc_conf = "xmalloc:true,lg_chunk:18,lg_tcache_max:12,purge:ratio";
 //const char *malloc_conf = "prof_leak:true,lg_prof_sample:0,prof_final:true,xmalloc:true,lg_tcache_max:16";
 
 int listen_fd;