This paper was presented at the 2009 International Symposium on Low Power Electronics and Design (ISLPED); the conference’s themes were green data centers and green computing. According to Meza et al., this paper is a first attempt at characterizing the power use of decision support systems, which is becoming an important concern for information technology (IT) executives.
The authors divide the system’s power use into several subsystems that use significant power. Interestingly, the power use of the disks and the “disk miscellaneous” subsystem amounts to over 57 percent of the system’s total power. The authors used Watts Up? power meters to measure the five subsystems; they range in power use from around five to seven percent for memory, to over 40 percent for the disks.
In order to allow measurements to be repeated, the authors chose a system configured to a “performance-optimized, audited TPC-H system at the 300GB scale.” TPC-H is a database benchmark for online ad hoc queries established by the Transaction Processing Performance Council (TPC); it closely resembles the typical use of online queries associated with the decision support systems for which the authors want to optimize power.
Surprisingly, the disk subsystem was oversized for optimal performance of the total system. Meza et al. reduced the original 204 disks to 32 core disks, with little degradation of performance. This suggests that database design and software improvement should be considered along with power use. The other subsystems’ power can also be optimized, but the improvement is less significant. The paper includes 20 references.
This paper will interest data center management professionals who want to optimize their power use by considering software and database design that can improve both the efficiency of the query system and its power performance. Database designers and database software developers should also take heed of these approaches, in order to implement green computing requirements that are compatible with query efficiency.
