[Colloquium] This morning: Ranganathan/Dissertation Defense/8-23-04
Margaret Jaffey
margaret at cs.uchicago.edu
Mon Aug 23 09:40:37 CDT 2004
Just a reminder that Kavitha Ranganathan's dissertation defense will be
held this morning at 10:30.
------
Department of Computer Science/The University of Chicago
*** Dissertation Defense ***
Candidate: Kavitha Ranganathan
Defense Date: Monday, August 23, 2004
Time and Location: 10:30 a.m. in Ryerson 255
Thesis Title: Management of Storage and Compute Resources in Large
Distributed Communities
Abstract:
Data Grids seek to harness geographically distributed resources
(storage, compute and network) for large-scale data-intensive problems
such as those encountered in high energy physics, bioinformatics, and
other disciplines. These problems typically involve numerous, loosely
coupled jobs that access large data sets. Effective resource management
in such environments is challenging, because of a need to address a
variety of metrics and constraints (for example, resource utilization,
response time, global and local allocation policies) while dealing with
multiple, potentially independent sources of resources and jobs.
The focus of this dissertation is the design of efficient and robust
resource management techniques that are effective across a wide range
of Data Grid scales, topologies and problem classes. We identify
specific requirements for a Data-Grid resource manager and propose a
decentralized and scalable architecture based on these requirements.
The solution comprises of autonomous agents spread across the Grid that
dynamically adapt their strategy according to perceived environment and
workload characteristics.
We also study the nature of sharing in such environments to ascertain
whether independent users will be willing to unconditionally let others
use their resources. We find that providing incentives for cooperation
is a critical step for the eventual success of such systems. To this
end, we define and evaluate soft-incentive schemes that can be easily
implemented under the proposed architecture.
We then go on to develop a family of job scheduling and data movement
(replication) strategies that fit into the proposed architecture.
Instead of studying the two aspects separately, we evaluate various
combinations of job and data scheduling strategies, under a wide range
of workload patterns, Grid topologies and user behavior. Two evaluation
techniques were used: a discrete event simulator, which we built for
modeling wide-area collaborations and a Grid testbed of 28 sites spread
across the county. Our results suggest that decoupling data movement
and computation scheduling has significant advantages when compared to
the traditional approach of data movement being tightly coupled to job
execution. We also find that dynamically adapting the scheduling
strategy according to workload characteristics has considerable
performance advantages.
Candidate's Advisor: Prof. Ian Foster
------------------------------------------------------------------------
---------------------------------
A draft copy of Ms. Ranganathan's dissertation is available in Ry 161A.
------------------------------------------------------------------------
---------------------------------
--
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Margaret P. Jaffey margaret at cs.uchicago.edu
Department of Computer Science
Student Support Rep (Ry 161A) (773) 702-6011
The University of Chicago http://www.cs.uchicago.edu
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
More information about the Colloquium
mailing list