Measuring Performance on HT-Enabled Multi-Core: Advantages of a Thread-Oriented Approach

Improve Server Performance Monitoring Software Multi-core Programming: From a hardware point of view, multi-core processors are simply single-die physical packages that provide capabilities similar to traditional symmetric multiprocessing (SMP) machines, and inevitably inherit most of the performance monitoring problems a programmer typically faces while working with SMP.

---

Introduction
By: Sergey N. Zheltov and Stanislav V. Bratanov, Senior Software Engineers, Intel Corporation

From a hardware point of view, multi-core processors are simply single-die physical packages that provide capabilities similar to traditional symmetric multiprocessing (SMP) machines, and inevitably inherit most of the performance monitoring problems a programmer typically faces while working with SMP.

We will address such performance estimation problems and advocate a thread-oriented approach to performance monitoring as an efficient way of dealing with multi-threaded code. At the same time, we will also address this as a basis for solving the particular task of monitoring Hyper-Threading Technology (HT Technology)-enabled multi-core systems.

This article discusses several methods and provides solutions to the problems which cannot be solved without the notion of the actual thread layout over time, thread activity, and interaction.

The problem of monitoring the utilization of resources shared between multiple cores or CPUs is discussed in an example of measuring bus utilization. A solution based on per-thread monitoring is suggested, and compared with sampling-based approaches.