diff --git a/conf/tfe/tfe.conf b/conf/tfe/tfe.conf
index 76b4973..fa5d0bd 100644
--- a/conf/tfe/tfe.conf
+++ b/conf/tfe/tfe.conf
@@ -8,6 +8,8 @@ enable_cpu_affinity=0
 # the first  mask for acceptor thread
 # the others mask for worker   thread
 cpu_affinity_mask=1-9,10-12
+# LEAST_CONN = 0; ROUND_ROBIN = 1, default 1
+load_balance=1
 
 [kni]
 ip=192.168.100.1
diff --git a/platform/include/internal/proxy.h b/platform/include/internal/proxy.h
index 40223e8..92a64d6 100644
--- a/platform/include/internal/proxy.h
+++ b/platform/include/internal/proxy.h
@@ -80,6 +80,12 @@ struct tfe_proxy_accept_para
 	unsigned int keyring_id;
 };
 
+enum tfe_load_balance_algo
+{
+	LEAST_CONN = 0,
+	ROUND_ROBIN = 1,
+};
+
 struct tfe_proxy
 {
 	char name[TFE_SYMBOL_MAX];
@@ -128,6 +134,9 @@ struct tfe_proxy
 	/* cpu affinity */
 	unsigned int enable_cpu_affinity;
 	unsigned int cpu_affinity_mask[TFE_THREAD_MAX];
+
+	/* load balancing */
+	enum tfe_load_balance_algo load_balance;
 };
 
 extern struct tfe_proxy * g_default_proxy;
diff --git a/platform/src/proxy.cpp b/platform/src/proxy.cpp
index 2887928..9d814a4 100644
--- a/platform/src/proxy.cpp
+++ b/platform/src/proxy.cpp
@@ -93,19 +93,34 @@ static __attribute__((__used__)) const char * tfe_version = "Unknown";
 struct tfe_thread_ctx * tfe_proxy_thread_ctx_acquire(struct tfe_proxy * ctx)
 {
 	unsigned int min_thread_id = 0;
+	unsigned int min_load;
 	static unsigned int counter=0;
 	counter++;
-	/*
-	for (unsigned int tid = 0; tid < ctx->nr_work_threads; tid++)
-	{
-		struct tfe_thread_ctx * thread_ctx = ctx->work_threads[tid];
-		unsigned int thread_load = ATOMIC_READ(&thread_ctx->load);
 
-		min_thread_id = min_load > thread_load ? tid : min_thread_id;
-		min_load = min_load > thread_load ? thread_load : min_load;
+	// least_conn
+	if (ctx->load_balance == LEAST_CONN)
+	{
+		for (unsigned int tid = 0; tid < ctx->nr_work_threads; tid++)
+		{
+			struct tfe_thread_ctx * thread_ctx = ctx->work_threads[tid];
+			unsigned int thread_load = ATOMIC_READ(&thread_ctx->load);
+			if (tid == 0)
+			{
+				min_thread_id = tid;
+				min_load = thread_load;
+				continue;
+			}
+
+			min_thread_id = min_load > thread_load ? tid : min_thread_id;
+			min_load = min_load > thread_load ? thread_load : min_load;
+		}
 	}
-*/
-	min_thread_id=counter%ctx->nr_work_threads;
+	// round_robin
+	else
+	{
+		min_thread_id = counter % ctx->nr_work_threads;
+	}
+	
 	ATOMIC_INC(&ctx->work_threads[min_thread_id]->load);
 	return ctx->work_threads[min_thread_id];
 }
@@ -308,6 +323,8 @@ int tfe_proxy_config(struct tfe_proxy * proxy, const char * profile)
 	MESA_load_profile_uint_def(profile, "system", "buffer_output_limit", &proxy->buffer_output_limit, 0);
 	MESA_load_profile_uint_def(profile, "system", "enable_cpu_affinity", &proxy->enable_cpu_affinity, 0);
 	MESA_load_profile_uint_range(profile, "system", "cpu_affinity_mask", TFE_THREAD_MAX, proxy->cpu_affinity_mask);
+	// LEAST_CONN = 0; ROUND_ROBIN = 1,
+	MESA_load_profile_uint_def(profile, "system", "load_balance", (unsigned int *)&proxy->load_balance, ROUND_ROBIN);
 
 	if (proxy->nr_work_threads < 1 || proxy->nr_work_threads > TFE_THREAD_MAX)
 	{