Events & News
Computer Science Colloquium
| Category | Lecture |
| Date | Thursday, April 3, 2008 |
| Time | 11:00 am |
| Location | GS 906 |
| Details | Light refreshments served in the 9th floor atrium at 10:45 AM. |
| Speaker | Song Jiang |
| Affiliation | Wayne State University |
Effective Provisioning of Consolidated Storage Service Using Machine-Learning Techniques
While today's computation becomes increasingly data centric, more and more applications and services rely on high-performance and reliable storage systems to deliver their promised quality services. Typical examples include Internet search engines, financial services, analysis of geno-medical data, and data mining for abstracting knowledge from massive amount of raw data. This technical trend has put a high demand on hardware and management investments on the storage systems, which may be beyond the resources available to a single division or department. Therefore, an increasingly common practice is to consolidate of storage resources into a data center to simultaneously provide storage services through the high-speed network to participating users. In addition, a consolidation of storage also provides high utilization of shared resource, ease of centralized management, and lower operating costs. In such a storage infrastructure, each user essentially reserves a virtual storage with contractual quality of services (QoS).
There are several critical challenges in effectively adopting the consolidation approach. The first is how to express the condition about users’ I/O requests under which QoS requirements should be honored. The second is how to enforce performance isolation so that a surge of resource demand from one user would not affect other users. The third is how to minimize the interference incurred by frequent ‘context switches’ between multiple virtual storages without violating QoS requirements. The last is to how to schedule streams of requests to different virtual storages with distinct QoS preferences, such as latency-sensitive or throughput-sensitive, so that the virtual storages can still be mutual beneficial to each other while they compete for resources of the same physical system.
Our work addresses these challenges by using several innovative techniques, including leveraging machine-learning techniques to estimate I/O request costs on reference systems, I/O-service-time-based resource allocations to accommodate various spatial localities, disk-oriented scheduling of grouped I/O requests for high disk efficiency, and symbiotic coordination of multiple virtual storages. Our experiment results have shown the effectiveness of the techniques in improving consolidated storage service for higher performance assurance, isolation, and system utilization.
Biography
Dr. Song Jiang is an assistant professor at the ECE department of Wayne State University. He received his Ph.D in computer science at the College of William and Mary in 2004. After that he had been a postdoctoral researcher at Los Alamos National Laboratory for two years. His current research projects include the provisioning of storage resources, middleware/compiler-level I/O arrangement, and management of buffer cache for high scalability in multi-processor/core systems. Several of his proposed algorithms, such as Swap-token and Clock-pro, have been incorporated into the official versions of Linux and NetBSD kernels.