• 版本、环境信息：

   1）PaddlePaddle版本：1.4.1
   2）CPU：CPU
   3）系统环境：centos

• 问题描述：使用ParallelExecutor进行infer，CPU利用率较低，和直接使用Executor的利用率差不多，速度没有明显提升，主要是想用ParallelExecutor数据并行提升预测速度

预测代码如下：

infer_program = fluid.default_main_program().clone(for_test=True)

    # For faster executor
    exec_strategy = fluid.ExecutionStrategy()
    exec_strategy.use_experimental_executor = True
    build_strategy = fluid.BuildStrategy()
    cpu_num = int(os.environ.get('CPU_NUM', cpu_count()))
    build_strategy.reduce_strategy = \
        fluid.BuildStrategy.ReduceStrategy.Reduce if cpu_num > 1 \
        else fluid.BuildStrategy.ReduceStrategy.AllReduce

    logger.info("begin executor")
    pe = fluid.ParallelExecutor(
        use_cuda=False,
        main_program=infer_program,
        build_strategy=build_strategy,
        exec_strategy=exec_strategy)

    fluid.io.load_persistables(exe, args.model_output_dir + '/batch-%d' % args.model_steps, infer_program)
    print 'start test'
    batch_id = 0
    py_reader.start()
    auc_metric = fluid.metrics.Auc("ROC")
    while True:
        try:
            results, y_pred, y_true = pe.run(fetch_list=[cost, preds, label])
            auc_metric.update(np.r_[y_pred], np.r_[y_true])
            results = np.mean(results)
            if batch_id % 100 == 0:
                sample = (batch_id + 1) * args.batch_size * int(os.environ.get('CPU_NUM', cpu_num))
                logger.info("TEST --> samples: {} batch: {} loss:{} auc:{}".format(sample, batch_id, results, auc_metric.eval()))
            batch_id += 1
        except fluid.core.EOFException:
            py_reader.reset()
            break

之前提的issue主题是运行出错发现是我自己代码写错了