Performance impact of resource contention in multicore systems

Abstract

Resource sharing in commodity multicore processors can have a significant impact on the performance of production applications. In this paper we use a differential performance analysis methodology to quantify the costs of contention for resources in the memory hierarchy of several multicore processors used in high-end computers. In particular, by comparing runs that bind MPI processes to cores in different patterns, we can isolate the effects of resource sharing. We use this methodology to measure how such sharing affects the performance of four applications of interest to NASA-OVERFLOW, MITgcm, Cart3D, and NCC. We also use a subset of the HPCC benchmarks and hardware counter data to help interpret and validate our findings. We conduct our study on high-end computing platforms that use four different quadcore microprocessors-Intel Clovertown, Intel Harpertown, AMD Barcelona, and Intel Nehalem-EP. The results help further our understanding of the requirements these codes place on their production environments and also of each computer's ability to deliver performance.

Measuring interference between live datacenter applications

“…The loosest interference class is between two applications which are on the same chip, but which do not share any resources except their interface to peripheral devices. Others have used interference classes to estimate the potential amount of interference in various assignments of applications to a machine (see contention groups in [19] for example). We see a few additional reasons that defining interference classes can be beneficial.…”

“…The works by Carter et al [9] and Levesque et al [30] evaluate whether increasing core counts on Cray machines will improve scientific applications' performance by estimating their memory bandwidth contention. Finally, Hood et al [19] and Jin et al [25] break down expected contention by class for different architectural platforms using microbenchmarks. They then estimate how real applications will perform on different architectural configurations.…”

Analyzing resource interdependencies in multi-core architectures to improve scheduling decisions

“…Jin and Hood et al [10,5] did similar studies and used an experimental setup like ours. However, they pursued the goal of identifying bottlenecks in HPC setups, so they evaluated MPI applications on fewer architectures and with less variations in the size of the data set compared to our work.…”

Validated Model-Based Performance Prediction of Multi-Core Software Routers

“…This situation is called resource contention which can significantly degrade the performance in comparison to a contentionfree environment. The effects of resource contention in multiprocessor and multi-core systems have been widely studied in the literature [30]- [32].…”