diff --git a/mindspore_serving/ccsrc/master/dispacther.cc b/mindspore_serving/ccsrc/master/dispacther.cc
index 92bf6e6..d99d786 100644
--- a/mindspore_serving/ccsrc/master/dispacther.cc
+++ b/mindspore_serving/ccsrc/master/dispacther.cc
@@ -55,15 +55,6 @@ DispatcherWorkerContext Dispatcher::GetWorkSession(const RequestSpec &request_sp
   return context;
 }
 
-void Dispatcher::Dispatch(const proto::PredictRequest &request, proto::PredictReply *reply) {
-  MSI_EXCEPTION_IF_NULL(reply);
-  auto promise = std::make_shared<std::promise<void>>();
-  auto future = promise->get_future();
-  PredictOnFinish on_finish = [promise]() { promise->set_value(); };
-  DispatchAsync(request, reply, on_finish);
-  future.get();  // wait callback finish
-}
-
 void Dispatcher::DispatchAsync(const proto::PredictRequest &request, proto::PredictReply *reply,
                                PredictOnFinish on_finish) {
   MSI_EXCEPTION_IF_NULL(reply);
diff --git a/mindspore_serving/ccsrc/master/dispacther.h b/mindspore_serving/ccsrc/master/dispacther.h
index 73f5235..15d8f5b 100644
--- a/mindspore_serving/ccsrc/master/dispacther.h
+++ b/mindspore_serving/ccsrc/master/dispacther.h
@@ -40,7 +40,6 @@ class MS_API Dispatcher {
  public:
   Dispatcher();
   ~Dispatcher();
-  void Dispatch(const proto::PredictRequest &request, proto::PredictReply *reply);
   void DispatchAsync(const proto::PredictRequest &request, proto::PredictReply *reply, PredictOnFinish on_finish);
 
   Status RegisterServable(const proto::RegisterRequest &request, proto::RegisterReply *reply);
diff --git a/mindspore_serving/ccsrc/master/restful/http_handle.cc b/mindspore_serving/ccsrc/master/restful/http_handle.cc
index 0a5d983..7812005 100644
--- a/mindspore_serving/ccsrc/master/restful/http_handle.cc
+++ b/mindspore_serving/ccsrc/master/restful/http_handle.cc
@@ -133,11 +133,12 @@ size_t GetTailEqualSize(const std::string &str) {
   return count;
 }
 
-Status HandleRestfulRequest(const std::shared_ptr<RestfulRequest> &restful_request, json *const out_json) {
+Status HandleRestfulRequest(const std::shared_ptr<RestfulRequest> &restful_request) {
   Status status(SUCCESS);
   std::shared_ptr<Dispatcher> dispatcher_ = Server::Instance().GetDispatcher();
-  RestfulService restful_service(dispatcher_);
-  status = restful_service.RunRestful(restful_request, out_json);
+  auto restful_service = std::make_shared<RestfulService>(dispatcher_);
+  status = restful_service->RunRestful(restful_request, restful_service);
+
   return status;
 }
 }  // namespace serving
diff --git a/mindspore_serving/ccsrc/master/restful/http_handle.h b/mindspore_serving/ccsrc/master/restful/http_handle.h
index bfbc318..78a8f74 100644
--- a/mindspore_serving/ccsrc/master/restful/http_handle.h
+++ b/mindspore_serving/ccsrc/master/restful/http_handle.h
@@ -25,7 +25,7 @@
 using nlohmann::json;
 namespace mindspore {
 namespace serving {
-Status HandleRestfulRequest(const std::shared_ptr<RestfulRequest> &restful_request, json *const out_json);
+Status HandleRestfulRequest(const std::shared_ptr<RestfulRequest> &restful_request);
 
 size_t Base64Encode(const uint8_t *input, size_t length, uint8_t *output);
 size_t Base64Decode(const uint8_t *target, size_t target_length, uint8_t *origin);
diff --git a/mindspore_serving/ccsrc/master/restful/http_process.cc b/mindspore_serving/ccsrc/master/restful/http_process.cc
index c99b179..c22f181 100644
--- a/mindspore_serving/ccsrc/master/restful/http_process.cc
+++ b/mindspore_serving/ccsrc/master/restful/http_process.cc
@@ -676,35 +676,35 @@ Status RestfulService::GetScalarData(const json &js, size_t index, bool is_bytes
 
   return status;
 }
-
 // 2.main
-Status RestfulService::RunRestful(const std::shared_ptr<RestfulRequest> &restful_request, json *const out_json) {
-  PredictRequest request;
-  PredictReply reply;
-
-  MSI_TIME_STAMP_START(ParseRequest)
-  auto status = ParseRequest(restful_request, &request);
-  MSI_TIME_STAMP_END(ParseRequest)
+Status RestfulService::RunRestful(const std::shared_ptr<RestfulRequest> &restful_request,
+                                  const std::shared_ptr<RestfulService> &restful_service) {
+  MSI_TIME_STAMP_START(RunRestful)
+  auto status = ParseRequest(restful_request, &request_);
   if (status != SUCCESS) {
     std::string error_msg = status.StatusMessage();
     std::string msg = "Parser request failed, " + error_msg;
+    nlohmann::json js;
+    js["error_msg"] = msg;
+    restful_request->RestfulReplay(js.dump());
     status = msg;
     return status;
   }
-
-  MSI_TIME_STAMP_START(Predict)
-  dispatcher_->Dispatch(request, &reply);
-  MSI_TIME_STAMP_END(Predict)
-
-  MSI_TIME_STAMP_START(CreateReplyJson)
-  status = ParseReply(reply, out_json);
-  MSI_TIME_STAMP_END(CreateReplyJson)
-  if (status != SUCCESS) {
-    std::string error_msg = status.StatusMessage();
-    std::string msg = "Parse reply failed, " + error_msg;
-    status = msg;
-    return status;
-  }
+  auto callback = [restful_service, restful_request, time_start_RunRestful]() {
+    nlohmann::json predict_json;
+    auto status = restful_service->ParseReply(restful_service->reply_, &predict_json);
+    if (status != SUCCESS) {
+      std::string error_msg = status.StatusMessage();
+      std::string msg = "Parse reply failed, " + error_msg;
+      nlohmann::json js;
+      js["error_msg"] = msg;
+      restful_request->RestfulReplay(js.dump());
+    } else {
+      restful_request->RestfulReplay(predict_json.dump());
+    }
+    MSI_TIME_STAMP_END(RunRestful)
+  };
+  dispatcher_->DispatchAsync(request_, &reply_, callback);
   return SUCCESS;
 }
 
diff --git a/mindspore_serving/ccsrc/master/restful/http_process.h b/mindspore_serving/ccsrc/master/restful/http_process.h
index 5e0f83a..fb0d6a5 100644
--- a/mindspore_serving/ccsrc/master/restful/http_process.h
+++ b/mindspore_serving/ccsrc/master/restful/http_process.h
@@ -45,7 +45,8 @@ class RestfulService {
  public:
   explicit RestfulService(const std::shared_ptr<Dispatcher> &dispatcher) : dispatcher_(dispatcher) {}
   ~RestfulService() = default;
-  Status RunRestful(const std::shared_ptr<RestfulRequest> &restful_request, json *const out_json);
+  Status RunRestful(const std::shared_ptr<RestfulRequest> &restful_request,
+                    const std::shared_ptr<RestfulService> &restful_service);
 
  private:
   Status CheckObjTypeMatchShape(DataType data_type, const std::vector<int64_t> &shape);
@@ -104,6 +105,8 @@ class RestfulService {
   int64_t instances_nums_{0};
   std::shared_ptr<Dispatcher> dispatcher_;
   std::vector<std::string> request_type_list_ = {kInstancesRequest};
+  proto::PredictRequest request_;
+  proto::PredictReply reply_;
 };
 
 }  // namespace serving
diff --git a/mindspore_serving/ccsrc/master/restful/restful_server.cc b/mindspore_serving/ccsrc/master/restful/restful_server.cc
index fd9c91d..31a4a6e 100644
--- a/mindspore_serving/ccsrc/master/restful/restful_server.cc
+++ b/mindspore_serving/ccsrc/master/restful/restful_server.cc
@@ -22,20 +22,7 @@
 namespace mindspore::serving {
 
 Status RestfulServer::Run(const std::shared_ptr<RestfulRequest> &restful_request) {
-  Status status(SUCCESS);
-  nlohmann::json predict_json;
-  std::string predict_str;
-  status = HandleRestfulRequest(restful_request, &predict_json);
-  if (status != SUCCESS) {
-    predict_str = status.StatusMessage();
-    nlohmann::json js;
-    js["error_msg"] = predict_str;
-    predict_str = js.dump();
-  } else {
-    predict_str = predict_json.dump();
-  }
-  restful_request->RestfulReplay(predict_str);
-  return status;
+  return HandleRestfulRequest(restful_request);
 }
 
 void RestfulServer::Committer(const std::shared_ptr<RestfulRequest> &restful_request) {