Departmental Colloquium

Some Empirical Experiences using Storage Class Memory: Evidence of Energy Efficiency, High Performance, and a Greener World

Professor Sam H. Noh

School of Computer and Information Engineering, Hongik University, Seoul, Korea

Thursday October 15, 2009
12:00 pm - SENSQ 5317

Hosted by Sangyeun Cho

Abstract

Storage Class Memory (SCM) is memory technology that blurs the distinction between memory and storage. With SCM, storage, which is traditionally a block device, now becomes byte-addressable; and memory, which is traditionally volatile, now becomes non-volatile. In this talk, we present results of some empirical studies that we have conducted with SCM. Though interest in SCM, in particular, Phase-Change Memory (PCM), has been growing, SCM is still a relatively new topic, especially in regards to system software. Hence, in the first part of this three-part talk, we will give a brief review of SCM and some recent work regarding SCM that will set a common ground for the next two parts of the talk.

In the second part of the talk, we address some emerging questions regarding the use of SCM. The answers that we present to the questions posed are based on empirical studies done on an embedded platform that employs FeRAM, a type of SCM. First, tradeoff between performance and energy efficiency when SCM comes into the memory hierarchy is examined. Through simple measurements for particular configurations, we show that by using FeRAM, power efficiency improves by a factor of 8, while performance is degraded only by a factor of 2.5 even for memory intensive applications. Second, and more importantly, we ask whether there are policies and mechanisms that we need to reconsider at the operating system level such that one can make use of both the non-volatility and random access features of SCM, deviating from the conventional usage of SCM as a simple replacement of RAM or storage. As answers to this question, schemes for metadata in-place update and single object management are proposed and exploited to reduce memory copies when manipulating file and memory objects. We show that through the use of these schemes, performance can be better than conventional systems despite the performance limitations of current SCM. Finally, we ask whether there might be features that traditional computing systems could not offer, but might be possible with SCM. As an answer to this question, we present a notion of persistent computing that supports lightweight system on/off capabilities by saving/restoring system contexts on SCM. This novel feature leads us to the last part of the talk.

In the third part of the talk, we take the notion of persistent computing to data centers. Studies have shown much of today's data centers are over-provisioned and underutilized. Over-provisioning cannot be avoided as these centers must anticipate peak load with bursty behavior. Under-utilization, to date, has also been unavoidable as systems always had to be ready for that sudden burst of requests that loom just around the corner. Previous research has pointed to turning off idle systems as one solution, albeit, an infeasible one due to its irresponsiveness. We show that with persistent computing through SCM, turning systems on and off can be achieved with minimum overhead. This feature is used to control systems on the whole(in comparison to previous fine-grained component-wise control) in fine time scale for high responsiveness for minimized power loss due to idleness. We present results of an empirical feasibility study that executes "real trace"-like workloads on a proof of-concept prototype "data center" composed of the aforementioned system. We quantify the energy savings and performance trade-off by turning idle systems off. We show that our energy savings approach consumes energy in proportion to user requests with configurable service of quality.

Biography of Speaker

Sam H. Noh
Department of Computer Engineering
Hongik University
Mapo Gu Sangsoo Dong 72-1
Seoul, Korea
+82-2-320-1470
samhnoh@hongik.ac.kr

I. Personal Information

Sam H. Noh
Full Professor
School of Information and Computer Engineering
Hongik University, Seoul Korea
Educational Background
B.S. Seoul National University (College of Engineering), Seoul, Korea 1986
(Computer Engineering)
Ph.D. University of Maryland, College Park, Maryland 1993
(Computer Science)

Employment Background

Hongik University, Assistant, Associate, Full Professor 1994.8~
University of Maryland, College Park 2001.8.27-2002.8.26
UMIACS, Visiting Associate Professor
George Washington University, Visiting Assistant Professor 1993.9.1-1994.6.30
Department of Statistics/Computer and Information Systems

Research Interest

Operating system issues and application of new memory technology such as Flash memory and
Storage Class Memory