diff --git a/Makefile b/Makefile
index d67e9852..7092729e 100644
--- a/Makefile
+++ b/Makefile
@@ -60,8 +60,8 @@ service/snlua.so : service-src/service_lua.c
 service/gate.so : gate/mread.c gate/ringbuffer.c gate/main.c
 	gcc $(CFLAGS) $(SHARED) $^ -o $@ -Igate -Iskynet-src
 
-luaclib/skynet.so : lualib-src/lua-skynet.c lualib-src/lua-seri.c lualib-src/lua-remoteobj.c | luaclib
-	gcc $(CFLAGS) $(SHARED) -Iluacompat $^ -o $@ -Iskynet-src
+luaclib/skynet.so : lualib-src/lua-skynet.c lualib-src/lua-seri.c lualib-src/lua-remoteobj.c lualib-src/trace_service.c | luaclib
+	gcc $(CFLAGS) $(SHARED) -Iluacompat $^ -o $@ -Iskynet-src -Ilualib-src
 
 service/client.so : service-src/service_client.c
 	gcc $(CFLAGS) $(SHARED) $^ -o $@ -Iskynet-src
diff --git a/lualib-src/lua-skynet.c b/lualib-src/lua-skynet.c
index 51267737..eaf97b52 100644
--- a/lualib-src/lua-skynet.c
+++ b/lualib-src/lua-skynet.c
@@ -1,4 +1,5 @@
 #include "skynet.h"
+#include "trace_service.h"
 #include "lua-seri.h"
 
 #include "luacompat52.h"
@@ -10,13 +11,12 @@
 #include <assert.h>
 #include <time.h>
 
-#define NANOSEC 1000000000
-
 struct stat {
 	lua_State *L;
 	int count;
 	uint32_t ti_sec;
 	uint32_t ti_nsec;
+	struct trace_pool *trace;
 };
 
 static void
@@ -25,23 +25,9 @@ _stat_begin(struct stat *S, struct timespec *ti) {
 	clock_gettime(CLOCK_THREAD_CPUTIME_ID, ti);
 }
 
-static void
+inline static void
 _stat_end(struct stat *S, struct timespec *ti) {
-	struct timespec end;
-	clock_gettime(CLOCK_THREAD_CPUTIME_ID, &end);
-	int diffsec = end.tv_sec - ti->tv_sec;
-	assert(diffsec>=0);
-	int diffnsec = end.tv_nsec - ti->tv_nsec;
-	if (diffnsec < 0) {
-		--diffsec;
-		diffnsec += NANOSEC;
-	}
-	S->ti_nsec += diffnsec;
-	if (S->ti_nsec > NANOSEC) {
-		++S->ti_sec;
-		S->ti_nsec -= NANOSEC;
-	}
-	S->ti_sec += diffsec;
+	diff_time(ti, &S->ti_sec, &S->ti_nsec);
 }
 
 static int
@@ -61,6 +47,10 @@ _stat(lua_State *L) {
 		lua_pushnumber(L, t);
 		return 1;
 	}
+	if (strcmp(what,"trace")==0) {
+		lua_pushlightuserdata(L, S->trace);
+		return 1;
+	}
 	return 0;
 }
 
@@ -111,6 +101,13 @@ _cb(struct skynet_context * context, void * ud, int type, int session, uint32_t
 	return 0;
 }
 
+static int
+_delete_stat(lua_State *L) {
+	struct stat * S = lua_touserdata(L,1);
+	trace_release(S->trace);
+	return 0;
+}
+
 static int
 _callback(lua_State *L) {
 	struct skynet_context * context = lua_touserdata(L, lua_upvalueindex(1));
@@ -128,6 +125,12 @@ _callback(lua_State *L) {
 	struct stat * S = lua_newuserdata(L, sizeof(*S));
 	memset(S, 0, sizeof(*S));
 	S->L = gL;
+	S->trace = trace_create();
+
+	lua_createtable(L,0,1);
+	lua_pushcfunction(L, _delete_stat);
+	lua_setfield(L,-1,"__gc");
+	lua_setmetatable(L, -2);
 
 	lua_rawsetp(L, LUA_REGISTRYINDEX, _stat);
 
@@ -337,6 +340,51 @@ _harbor(lua_State *L) {
 	return 2;
 }
 
+// trace api
+static int
+_trace_new(lua_State *L) {
+	struct trace_pool *p = lua_touserdata(L,1);
+	struct trace_info *t = trace_new(p);
+	lua_pushlightuserdata(L,t);
+	return 1;
+}
+
+static int
+_trace_delete(lua_State *L) {
+	struct trace_pool *p = lua_touserdata(L,1);
+	struct trace_info *t = lua_touserdata(L,2);
+	double ti = trace_delete(p,t);
+	lua_pushnumber(L, ti);
+	return 1;
+}
+
+static int
+_trace_switch(lua_State *L) {
+	struct trace_pool *p = lua_touserdata(L,1);
+	int session = luaL_checkinteger(L,2);
+	trace_switch(p, session);
+	return 0;
+}
+
+static int
+_trace_yield(lua_State *L) {
+	struct trace_pool *p = lua_touserdata(L,1);
+	struct trace_info * t = trace_yield(p);
+	if (t) {
+		lua_pushlightuserdata(L,t);
+		return 1;
+	}
+	return 0;
+}
+
+static int
+_trace_register(lua_State *L) {
+	struct trace_pool *p = lua_touserdata(L,1);
+	int session = luaL_checkinteger(L,2);
+	trace_register(p, session);
+	return 0;
+}
+
 // define in lua-remoteobj.c
 int remoteobj_init(lua_State *L);
 
@@ -379,10 +427,14 @@ luaopen_skynet_c(lua_State *L) {
 	luaL_Reg l2[] = {
 		{ "stat", _stat },
 		{ "remote_init", remoteobj_init },
+		{ "trace_new", _trace_new },
+		{ "trace_delete", _trace_delete },
+		{ "trace_switch", _trace_switch },
+		{ "trace_yield", _trace_yield },
+		{ "trace_register", _trace_register },
 		{ NULL, NULL },
 	};
 
-
 	luaL_setfuncs(L,l2,0);
 
 	return 1;
diff --git a/lualib-src/trace_service.c b/lualib-src/trace_service.c
new file mode 100644
index 00000000..f6040581
--- /dev/null
+++ b/lualib-src/trace_service.c
@@ -0,0 +1,146 @@
+#include "trace_service.h"
+#include <stdlib.h>
+#include <string.h>
+#include <assert.h>
+#include <time.h>
+#include <stdint.h>
+
+#define HASH_SIZE 32
+
+void 
+diff_time(struct timespec *ti, uint32_t *sec, uint32_t *nsec) {
+	struct timespec end;
+	clock_gettime(CLOCK_THREAD_CPUTIME_ID, &end);
+	int diffsec = end.tv_sec - ti->tv_sec;
+	assert(diffsec>=0);
+	int diffnsec = end.tv_nsec - ti->tv_nsec;
+	if (diffnsec < 0) {
+		--diffsec;
+		diffnsec += NANOSEC;
+	}
+	*nsec += diffnsec;
+	if (*nsec > NANOSEC) {
+		++*sec;
+		*nsec -= NANOSEC;
+	}
+	*sec += diffsec;
+}
+
+struct trace_info {
+	int session;
+	struct trace_info * prev;
+	struct trace_info * next;
+	struct timespec ti;
+	uint32_t ti_sec;
+	uint32_t ti_nsec;
+};
+
+struct trace_pool {
+	struct trace_info * current;
+	struct trace_info * slot[HASH_SIZE];
+};
+
+struct trace_pool *
+trace_create() {
+	struct trace_pool * p = malloc(sizeof(*p));
+	memset(p, 0, sizeof(*p));
+	return p;
+}
+
+static void
+_free_slot(struct trace_info *t) {
+	while (t) {
+		struct trace_info *next = t->next;
+		free(t);
+		t = next;
+	}
+}
+
+void
+trace_release(struct trace_pool *p) {
+	int i;
+	for (i=0;i<HASH_SIZE;i++) {
+		_free_slot(p->slot[i]);
+	}
+	free(p->current);
+}
+
+struct trace_info *
+trace_new(struct trace_pool *p) {
+	assert(p->current == NULL);
+	struct trace_info *t = malloc(sizeof(*t));
+	p->current = t;
+	t->session = 0;
+	t->prev = NULL;
+	t->next = NULL;
+	t->ti_sec = 0;
+	t->ti_nsec = 0;
+	clock_gettime(CLOCK_THREAD_CPUTIME_ID, &t->ti);
+	return t;
+}
+
+void 
+trace_register(struct trace_pool *p, int session) {
+	struct trace_info *t = p->current;
+	int hash = session % HASH_SIZE;
+	assert(t->session == 0);
+	t->session = session;
+	t->prev = NULL;
+	t->next = p->slot[hash];
+	if (p->slot[hash]) {
+		p->slot[hash]->prev = t;
+	}
+}
+
+struct trace_info * 
+trace_yield(struct trace_pool *p) {
+	struct trace_info *t = p->current;
+	if (t == NULL)
+		return NULL;
+	p->current = NULL;
+
+	diff_time(&t->ti,&t->ti_sec,&t->ti_nsec);
+
+	if (t->session == 0) {
+		return t;
+	} else {
+		return NULL;
+	}
+}
+
+void 
+trace_switch(struct trace_pool *p, int session) {
+	assert(p->current == NULL);
+	struct trace_info *t = p->slot[session % HASH_SIZE];
+	struct trace_info *prev = NULL;
+	while (t) {
+		if (t->session == session) {
+			p->current = t;
+			t->session = 0;
+			if (prev == NULL) {
+				p->slot[session % HASH_SIZE] = t->next;
+			} else {
+				prev->next = t->next;
+			}
+			if (t->next) {
+				t->next->prev = prev;
+			}
+			return;
+		}
+		prev = t;
+		t=t->next;
+	}
+}
+
+double 
+trace_delete(struct trace_pool *p, struct trace_info *t) {
+	assert(p->current == t);
+	p->current = NULL;
+	if (t) {
+		double ti = (double)t->ti_sec + (double)t->ti_nsec / NANOSEC;
+		free(t);
+		return ti;
+	} else {
+		return 0;
+	}
+}
diff --git a/lualib-src/trace_service.h b/lualib-src/trace_service.h
new file mode 100644
index 00000000..61fbc79a
--- /dev/null
+++ b/lualib-src/trace_service.h
@@ -0,0 +1,22 @@
+#ifndef TRACE_SERVICE_H
+#define TRACE_SERVICE_H
+
+#include <time.h>
+#include <stdint.h>
+
+#define NANOSEC 1000000000
+
+void diff_time(struct timespec *ti, uint32_t *sec, uint32_t *nsec);
+
+struct trace_pool;
+struct trace_info;
+
+struct trace_pool *  trace_create();
+void trace_release(struct trace_pool *);
+struct trace_info * trace_new(struct trace_pool *);
+void trace_register(struct trace_pool *, int session);
+void trace_switch(struct trace_pool *, int session);
+struct trace_info * trace_yield(struct trace_pool *);
+double trace_delete(struct trace_pool *, struct trace_info *);
+
+#endif