Differences

This shows you the differences between two versions of the page.

Link to this comparison view

— seminars:seminar_5_12_17 [2017/09/21 02:02] (current)
Line 1: Line 1:
 +====== Network-driven Context-Aware Datacenter Server Management: Hardware/​Software Cross-Layer Approaches ======
  
 +Friday May 12, 2017\\
 +Location: CIC Panther Hollow Room\\
 +Time: 11:00PM\\
 +
 +
 +
 +**[[https://​www.ece.illinois.edu/​directory/​profile/​nskim|Nam Sung Kim (UIUC)]]**\\
 +
 +=====Abstract=====
 +Datacenter servers running latency-critical online applications such as web
 +search and social network are often underutilized as they are
 +(over)provisioned for unpredictable peak service demand. This in turn hurts
 +energy efficiency and thus the total cost of ownership (TCO) due to poor
 +energy proportionality of contemporary servers. To improve energy
 +efficiency, we may consider to deploy aggressive power management policies
 +and run latency-critical applications with latency-agnostic,​ throughput
 +applications. Such approaches, however, prone to violations of service-level
 +agreement (SLA) whenever servers need to change performance and power
 +states, which prevents servers from performing any tasks for a notable
 +amount of time, and co-running throughput applications generate block I/O
 +requests, which severely interferes with memory requests from
 +latency-critical applications at the memory subsystem.
 +
 +In this talk , tackling these challenges, I will present two techniques
 +dubbed NCAP and CLAUD. More specifically,​ NCAP and CLAUD exploit the fact
 +that the rate of network packets encapsulating requests from clients can
 +significantly affect the utilization of datacenter servers. Thus, we propose
 +to enhance network interface cards (NICs) and NIC drivers to measure the
 +rate of latency-critical network packets and send special interrupts to the
 +operating system when the rate exceeds certain values. The special
 +interrupts allow (1) NCAP to preemptively change performance and power
 +states of servers and (2) CLAUD to proactively throttle block I/O requests
 +from co-running throughput applications. NCAP and CLAUD in turn allow
 +datacenter operators to more aggressively deploy power management policies
 +and/or run latency-critical applications with throughput applications
 +without violating SLAs.
 +
 +
 +=====Bio=====
 +Nam Sung Kim is an IEEE Fellow and an Associate Professor at the University
 +of Illinois, Urbana-Champaign and. Prior to joining the University of
 +Illinois in the fall of 2015, He was an Associate Professor at the
 +University of Wisconsin, Madison where he was early-tenured in 2013. His
 +interdisciplinary research incorporates device, circuit, architecture,​ and
 +software for power-efficient computing. Prior to joining the University of
 +Wisconsin, Madison, he was a senior research scientist at Intel from 2004 to
 +2008, where he conducted research in power-efficient digital circuit and
 +process architecture. He has published nearly 160 refereed articles to
 +highly-selective conferences and journals in the field of digital circuit,
 +processor architecture,​ and computer-aided design. The top three most
 +frequently cited papers have more than 3500 citations and the total number
 +of citations of all his papers exceeds 6800. He is a recipient of the IEEE
 +Design Automation Conference (DAC) Student Design Contest Award in 2001,
 +Intel Fellowship in 2002, IEEE International Symposium on Microarchitecture
 +(MICRO) Best Paper Award in 2003, NSF CAREER Award in 2010, IBM Faculty
 +Award in 2011 and 2012, and  University of Wisconsin Villas Associates Award
 +in 2015. I am a member of IEEE International Symposium on High-Performance
 +Computer Architecture (HPCA) Hall of Fame and IEEE International Symposium
 +on Microarchitecture (MICRO) Hall of Fame.
 +
 +
 +\\
 +\\
 +**[[seminars| Back to the seminar page]]**