Architecture: Matthew Arnold/IBM T.J. Watson Research Center The Future of Virtual Machine Performance in ACES 2.302

Contact Name: 
Jenna Whitney
Feb 7, 2006 10:00am - 11:00am

Speaker Name/Affiliation: Matthew Arnold/IBM T.J.
Watson Research Center

Talk Title: The Future of Virtual Machine P


Date/Time: February 7 2006 at 10:00 a.m.

9:45 a.m.

Location: ACES 2.302

Host: Kathryn McKinley

br>Talk Abstract:
Users of virtual machines care most about two aspects

performance: startup and throughput. In this talk I will
give a
brief overview of the techniques commercial VMs use
to improve these a

spects of performance and discuss the
challenges that still remain. I

will then present two new
nontraditional approaches for making progres

s in these areas.

1) Improving startup performance using a cross-run
repository (OOPSLA''05). Despite the important role that

rofiling plays in achieving high performance current virtual

discard a program''s profile data at the end of
execution. Our work pre

sents a fully automated architecture
for exploiting cross-run profile d

ata in virtual machines.
This work addresses a number of challenges tha

t previously
limited the practicality of such an approach.

2) Th

roughput performance: Online Performance Auditing
(PLDI''06). This work
describes an online framework for evaluating
the effectiveness of opti

mizations enabling an online system
to automatically identify and corr

ect performance anomalies
that occur at runtime. This work encourages a
shift in the
way optimizations are developed and tuned for online syst

and may allow much of the work in offline empirical optimization <

br>search to be applied automatically at runtime.

All of this work i

s implemented and evaluated using IBM''s
product J9 Java Virtual Machin


Speaker Bio:
Matthew Arnold received his Ph.D. from Rutgers

in 2002 and is now a Research Staff Member at the IBM T.J.

Watson Research Center in Hawthorne NY. For his thesis
work he dev

eloped low-overhead profiling techniques and
showed how they can be use

d to drive feedback-directed optimization
in a virtual machine; this w

ork is currently used in IBM''s
product JVM. He has worked with the Jik

es Research Virtual
Machine and IBM''s production JVM and continues to
use both
for his research. His current interests include virtual

achine performance low overhead profiling and dynamic
analysis of sof