[dbjapan] 講演 Prof.Calton Pu 5/23 17:45-

[Masaru Kitsuregawa <kitsure [at] tkl.iis.u-tokyo.ac.jp>]

データベースな皆様へ

以下の講演会を開催致します。ふるってご参加ください。  喜連川優

 日時　平成２８年５月２３日（月）　　17時45分〜
 場所　東京大学生産技術研究所　As棟 中セミナー５(As313・314)
　　　　住所：東京都目黒区駒場4-6-1
              http://www.iis.u-tokyo.ac.jp/access/access.html
              http://www.iis.u-tokyo.ac.jp/access/campusmap.html
        問合せ先：東京大学生産技術研究所Ee310 喜連川教授室　03-5452-6254
                  secretaries [at] tkl.iis.u-tokyo.ac.jp>

----------------------

An Experimental Study of Latency Long Tail Problem: Impact of
Millibottlenecks in Cloud Environments

Calton Pu
Professor and John P. Imlay, Jr. Chair in Software School of Computer
Science, Georgia Institute of Technology


ABSTRACT

Large mission-critical n-tier applications running in data centers and
cloud environments have complex deployment dependencies and stringent
quality of service requirements. One of the most interesting questions
is the wide fluctuations of response time, often referred to as the
latency long tail problem. They happen even when the system utilization
is still very far from saturation (e.g., 60% average CPU utilization).
Using automated n-tier application benchmark deployment, monitoring, and
analysis, we have found several of these occasional response time
fluctuations (several seconds) that contribute to the latency long tail.
Our study presents convincing evidence that one of the causes comes from
a seemingly unlikely source: very short bottlenecks that last only
tens of milliseconds. These millibottlenecks represent an interesting
new class of performance phenomena for several reasons. First, they are
very short but have large impact (two orders of magnitude longer
response time delay than the duration of a millibottlenecks). Second,
they strike seemingly randomly (the long queries only take milliseconds
when executed by themselves).
Third, they can happen at moderate CPU utilization levels. We have
identified causes of millibottlenecks at several system layers,
including processor architecture (e.g., dynamic voltage and frequency
scaling ・DVFS),system software (e.g., Java VM garbage collection), and
VM-based application consolidation.


--------------------------
Bio Sketch: Calton Pu was born in Taiwan and grew up in Brazil.  He
received his PhD from University of Washington in 1986 and served on the
faculty of Columbia University and Oregon Graduate Institute.
Currently, he is holding the position of Professor and John P. Imlay,
Jr.Chair in Software in the College of Computing, Georgia Institute of
Technology.  He has worked on several projects in systems and database
research.  His contributions to systems research include program
specialization and software feedback.  His contributions to database
research include extended transaction models and their implementation.
His recent research has focused on automated system management in clouds
(Elba project), information quality (e.g., spam processing), and big
data in Internet of Things.  He has collaborated extensively with
scientists and industry researchers.  He has published more than 70
journal papers and book chapters, 280 conference and refereed workshop
papers. He served on more than 120 program committees, including the
co-PC chairs of SRDS'95, ICDE・9,COOPIS・2, SRDS・3, DOA・7, DEBS・9,
ICWS・0, CollaborateCom'11, ICAC・3, CLOUD・5, BigData Congress・6, and
co-general chair of ICDE'97, CIKM'01, ICDE・6, DEPSA・7, CEAS・7, SCC・
8, CollaborateCom・8, World Service Congress・1, CollaborateCom・2, IEEE
CIC・5 and CIC・6.