Power Measurements and Analyses of Massive Object Storage System

With the increase of high performance computing systems' scale, the power consumption of HPC systems greatly increase, among which the power consumption of massive storage system occupies an important role. Measurements and analyses of massive storage system are significant for reducing the power consumption of storage system. Tianhe-1 comes in at No.5 in Nov. 2009 Top500 list with the highest ranking ever for a Chinese supercomputer. In this article, we do actual measurement using WT1600 digital power meter for object storage system under different kinds of conditions, which belongs to the same series with the storage system in Tianhe-1. We discuss and analyze the measurement results and give some important and interesting conclusions. The detailed works include: 1) comparing the power characteristics of read/write operations; 2) comparing the power of different RAID levels; 3) comparing the power under different number of processes; 4) comparing the power with different power levels. These discussions are meaningful for guiding design and power optimization of the future massive object storage system.

I. Introduction

Too high power has been a serious problem of High Performance Computing (HPC) systems[1]. First, increasing frequency and performance boost bring about the power increase of single computing component. Second, with the scale of HPC systems increases, the power consumption of the whole system reaches a very high level. For example, in the Top500 list published in Nov. 2009[2], the average power of top 10 computers reaches 2000KW, among which the highest power has arrived at 7000 KW. Too high power will bring about negative effects on system stability, usability, maintainability and so on. Meanwhile, extra high power will increase the running cost and is harmful to the environments. Power consumption has been one of the important constraints for HPC system design.