Software-Based Disk Power Management for Scientific Applications
Open Access
- Author:
- Son, Seung Woo
- Graduate Program:
- Computer Science and Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- July 07, 2008
- Committee Members:
- Mahmut Taylan Kandemir, Committee Chair/Co-Chair
Padma Raghavan, Committee Member
Bhuvan Urgaonkar, Committee Member
Long Qing Chen, Committee Member - Keywords:
- Low power
compiler
disk subsystem
parallel file system
runtime - Abstract:
- Power consumption by high-performance systems is becoming an increasing concern for system designers and software writers a like. Disk subsystem is known to be a major contributor to the overall power budget of high-performance systems. Most scientific applications today rely heavily on disk I/O for out-of-core computations, checkpointing, and data visualization. To reduce excess energy consumption on disk system, prior studies proposed several hardware or OS-based disk power management schemes. While such schemes have been known to be effective in certain cases, they might miss opportunities for better energy savings due to their reactive nature. While compiler based schemes can make more accurate decisions on a given application by extracting disk access patterns statically, the lack of runtime information on the status of shared disks may lead to wrong decisions when multiple applications exercise the same set of disks concurrently. Therefore, quantitative comparison of reactive, compiler-based, and hybrid schemes is very important. This dissertation makes four major contributions towards more effective disk power management for scientific applications that use disk-resident data frequently (i.e., so called I/O-intensive applications). First, it shows that, while conventional hardware based disk power management scheme is useful in certain cases, compiler-driven approach can be more effective for array-based scientific applications executing on parallel architectures. Second, it shows that restructuring the application code increases length of disk idle periods, thereby leading to better exploitation of available power-saving capabilities. Third, it proposes a compiler-directed energy-aware prefetching scheme for scientific applications that process disk-resident data sets. Finally, it proposes a runtime system support for software-based disk power management scheme. The proposed runtime system is implemented within PVFS2, a parallel file system. We conclude by a brief discussion of ongoing and future work on I/O.