增加线程数的高延迟答案

【问题标题】：High latency on increasing thread count增加线程数的高延迟
【发布时间】：2016-07-24 02:56:32
【问题描述】：

在下面的代码中 DataGather = endDataGather - beginDataGather 需要 1.7ms & 服务响应时间 = service_COMPLETED - service_REQUEST_SENT 从 20us 到 200us 不等（因为它们在同一个局域网上被模拟为虚拟，因此如此之低）现在，如果我将 tomcat8 线程从 10 增加到 200，DataGather 增加到 150ms +，即使我将线程从 200 增加到 1000，它甚至会增加 250+。机器规格 8 核 Xenon，64gb ram。当 apache 基准测试使用 -n 40000 -c 100 args 运行时测量时间，这是由于线程调度/上下文切换还是其他原因？我如何摆脱这种变化？当延迟为 20-100 毫秒的真实服务出现时，它会保留吗？

       public List<ServiceResponse> getData(final List<Service> services, final Data data) {
          //beginDateGather;

          final List<ServiceResponse> serviceResponses = Collections.synchronizedList(new ArrayList<>());
          try {
            final CountDownLatch latch = new CountDownLatch(services.size());
            Map<Future<HttpResponse>, HttpRequestBase> responseRequestMap = new HashMap<Future<HttpResponse>, HttpRequestBase>();

            for (final service service : services) {
              //creating request for a service
              try {
                HttpRequestBase request = RequestCreator.getRequestBase(service, data);
                //service_REQUEST_SENT
                Future<HttpResponse> response = client.execute(request,
                    new MyFutureCallback(service, data, latch, serviceResponses));
                responseRequestMap.put(response, request);
              } catch (Exception e) {
                latch.countDown();
              }
            }
            try {
              boolean isWaitIsOver = latch.await(timeout, TimeUnit.MILLISECONDS);
              if (!isWaitIsOver) {
                for (Future<HttpResponse> response : responseRequestMap.keySet()) {
                  if (!response.isDone()) {
                    response.cancel(true);
                  }
                }
              }
            } catch (InterruptedException e) {
            }
          } catch (Exception e) {
          }
          //endDataGather
          return serviceResponses;
    }


     public class MyFutureCallback implements FutureCallback<HttpResponse> {

        private Service service;
        private Data data;
        private CountDownLatch latch;
        private List<serviceResponse> serviceResponses;

        public MyFutureCallback( Service service, Data data, CountDownLatch latch, List<ServiceResponse> serviceResponses) {
          this.service = service;
          this.data = data;
          this.latch = latch;
          this.serviceResponses = serviceResponses;
        }

        @Override
        public void completed(HttpResponse result) {
          try {
            ServiceResponse serviceResponse = parseResponse(result, data, service);
              serviceResponses.add(serviceResponse);
          } catch (Exception e) {
          } finally {
            //service_COMPLETED
            latch.countDown();
          }
        }

        @Override
        public void failed(Exception ex) {
          latch.countDown();
        }

        @Override
        public void cancelled() {
          latch.countDown();
        }
       }

【问题讨论】：

标签： java multithreading apache-httpclient-4.x countdownlatch

【解决方案1】：

是的，这似乎是由于线程的上下文切换。在这种情况下，增加线程数无济于事。您可以使用线程池进行回调。检查此链接以供参考并尝试使用 .PoolingClientAsyncConnectionManager

How to use HttpAsyncClient with multithreaded operation?

【讨论】：