GRAM
(Grid Resource and Application Matching)
Sponsors: DOE
(via LACSI) , NSF
Research into ideas, algorithms
and systems to observe and measure application and system behavior, on
shared clusters and grid environments, and employ that knowledge
towards resource allocation and management. A summary of the work going
on in this project can be obtained from the following poster:
Tools and Middleware for Parallel and
Distributed Processing on Computation Grids.
Key components of ongoing research include the following:
- Automatic construction of "application performance
skeletons". A skeleton is a short running program that mimics the
execution of application it represents, and is used as a probe to
measure expected application behavior in an unpredictable network
environment.
- Modeling the network performance that application operations can
expect on shared networks, specifically collective comunication
operations.
The approach is based on synchronized physical network measurements and
modeling network sharing.
- Rapid estimation of TCP bandwidth based on inferring flow
status from segment arrivals.
- Exploring job scheduling models, such as gang scheduling ands
independent scheduling for shared environments.
Students currently active in this project:
Mark Huang , Sathish Kumar, Sukhdeep Sodhi , Aditya
Toomula, Deepti Vyas, Qiang
Xu
Recent publications most relevant to this project:
- S. Sodhi and J. Subhlok, Skeleton
Based Performance Prediction on Shared Networks, Grids and
Advanced Networks Workshop (GAN'04) at CCGRID 2004:
The IEEE International Symposium on Cluster Computing and the Grid,
Chicago, IL, April 2004. abstract
pdf
, postscript
- S. Venkataramaiah and J. Subhlok , Performance Estimation
for Scheduling on Shared Networks, 9th Workshop on Job Scheduling
Strategies for Parallel Processing, Seattle, WA, June 2003 pdf , postscript
, powerpoint
- S. Goteti and J. Subhlok, Communication Pattern Based
Node Selection for Shared Networks, Autonomic Computing Workshop:
The
Fifth Annual International Workshop on Active Middleware Services (AMS
2003),
Seattle, WA, June, 2003. abstract
, pdf , postscript , powerpoint
- A. Singh and J. Subhlok, Reconstruction of Application
Layer Message Sequences by Network Monitoring, IASTED International
Conference on Communications and Computer Networks 2002, Cambridge, MA,
November 2002 pdf , powerpoint
- S. Venkataramaiah and J. Subhlok Performance Prediction for
Simple CPU and Network Sharing, LACSI Symposium 2002, Santa Fe, New
Mexico, October 2002. pdf , postscript , powerpoint
- J. Subhlok, S. Venkataramaiah and A. Singh Characterizing
NAS Benchmark Performance on Shared Heterogeneous Networks. 11th
Heterogeneous Computing Workshop, (HCW '02), held with IPDPS '02, Fort
Lauderdale, April 2002. pdfabstract
, pdf , postscript ,
powerpoint
Please contact us for drafts of current work.
DANCE
(Distributed Adaptive Network Caching)
Sponsors: Texas
Higher Education Coordinating Board (via ATP program), Compaq/HP, iMimic
The goal of the this project is dynamic, adaptive distributed
caching of web objects.
Students currently active in this project:
Qiaomei Huang , Ming Zu
Recent publications most relevant to this project
- M. Zu and J. Subhlok, Home Based Cooperative Web Caching,
Seventh Multi-Conference on Systemics, Cybernetics and
Informatics, Orlando, FL,
July 2003. pdf
- P. Desai and J. Subhlok. An
Adaptive Hierarchy Management System for Web Caches, The 4th
International Conference on Internet Computing, Las Vegas, NV, June
2003. pdf , postscript , powerpoint
- P. Du and J. Subhlok. Evaluation
of Performance of Cooperative Web Caching with Web Polygraph,
7th
International Workshop on Web Content Caching and Distribution,
Boulder,
CO, July 2002. pdf , postscript
, powerpoint
Sponsor:
University of Houston, Educational Technology and Outreach ( FDIP B
Program)
This is an educational project. We are developing a comprehensive
interactive multimedia introduction to a selected set of courses
offered by the Computer Science department. course prospectus
includes a multimedia introduction to the syllabus, audio/video clips
from lectures, interactive quizzes to test
background knowledge, and animations to demonstrate key concepts and
problem
solving. This pilot project is intended to lead the way to
interactive multimedia prospectuses for courses and degree programs
across the University.