Warehouse-scale Computers: Opportunities and Challenges

Speaker:        Professor Mary Lou Soffa
                Department of Computer Science
                University of Virginia

Title:          "Warehouse-scale Computers: Opportunities and Challenges"

Date:           Monday, 6 January 2014

Time:           4:00pm - 5:00pm

Venue:          Room 4472 (via lifts 25/26), HKUST

Abstract:

Web-service companies such as Google, Microsoft, Amazon, Yahoo, and Apple
spend hundreds of millions of dollars to construct and operate
Warehouse-scale Computers (WSC) which provide popular web-services such as
search, social networking, webmail, video streaming, enterprise management
tools, online maps, automatic translation, and online courses. The primary
advantages of WSC are the scalability and cost benefits for both the
end-users and web-service companies. These WSCs house hundreds to
thousands of machines to provide the computing resources needed to serve
millions of users. To limit the cost of ownership of WSCs, these machines
are composed of commodity components which are cheap and easily
replaceable, often multi-cores. When multiple applications are running
simultaneously on a multi-core machine, resources sharing and contention
among cores can result in a significant amount of performance
interference. This interference leads to a significant problem in meeting
the requirements of user facing web-service applications. To avoid the
constant unpredictable threat that shared resource contention poses to an
application's QoS, datacenter operators typically disallow co-locations of
latency-sensitive jobs with other jobs. This unnecessary over-provisioning
of computer resources reduces the overall utilization of WSCs and results
in an unnecessarily high cost and a large environmental footprint for a
given set of web-service workloads. In this talk, I discuss these issues
and present our research using scheduling and compiling to improve the
capability and cost effectiveness by improving resource efficiency.
Specifically, we reconcile the apparent conflict between the need to
maintain high QoS for latency-critical, high-priority services and the
desire to increase hardware utilization by scheduling multiple workloads
per server.

*******************
Biography:

Mary Lou Soffa is the Owen R. Cheatham Professor of Sciences at the
Computer Science Department at the University of Virginia, serving as the
Department Chair from 2004 to 2012.  From 1977 to 2004, she was a
Professor of Computer Science at the University of Pittsburgh and also
served as the Dean of Graduate Studies in the College of Arts and
Sciences. Her research interests include software systems for multi-core
architectures, optimizing compilers, software testing, program analysis
and software security. She has directed 30 Ph.D. students to completion,
half of whom are women. Mary Lou received the Ken Kennedy Award in 2012
for contributions to compiler technology and software engineering,
exemplary service to the profession, and lifelong dedication to mentoring
and improving diversity in computing. She has directed 30 Ph.D. students
to completion, half of whom are women and two are minorities.  Mary Lou is
both an IEEE Fellow and an ACM Fellow. She received the Anita Borg
Technical Leadership Award and the Presidential Award for Excellence in
Science, Mathematics and Engineering Mentoring.  She received the
Computing Research Association (CRA) Nico Habermann Award in 2006 and was
selected as a Girl Scout Woman of Distinction in 2003. She has served as
conference chair, program chair or program committee member for numerous
conferences. She currently serves on the ACM Publication Board and ACM
Council.  She also serves on the CRA-W  Board and CRA-E Committee.