[Colloquium] Indranil Gupta - Wed. April 16, 2003

[Margery Ishmael]

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

DEPARTMENT OF COMPUTER SCIENCE - TALK

Wednesday, April 16, 2003 at 2:30 pm in Ryerson 251

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

Speaker: Indranil Gupta
From: Cornell University
http://www.cs.cornell.edu/gupta/

Title: Scaling Distributed Group Communication -- A Probabilistic Protocol 
Methodology

=============================

Abstract: For several years, peer-to-peer technology has been at the focus 
of increasing attention among users and researchers. It is used in the 
design of file sharing systems (e.g., Napster, Kazaa), data centers and 
distributed file systems. It is likely to influence evolution of 
intelligent, autonomous and large-scale systems of small devices (e.g., 
MEMS sensor nodes). The success of such systems depends on their ability to 
sustain performance under dynamic stresses (network congestion, end-point 
failures, system churn) as well as to scale gracefully when end-points 
number into thousands or millions. The traditional approaches to designing 
distributed protocols (e.g., two phase commit, heartbeating) have 
difficulty scaling up, and are restricted in some settings to a few hundred 
participants and very light ranges of perturbation. I will discuss an 
alternative methodology to design large-scale distributed protocols. This 
methodology generates probabilistic protocols for a suite of problems 
central to the design of large-scale peer-to-peer systems. We have designed 
and experimented with a group membership software module, a system for 
resource location and discovery, protocols for variants of reliable 
multicast and consensus, and a data aggregation protocol. Probabilistic 
protocols impose low overheads on the group (costs at participants are 
often constant), and offer low probabilities of incorrectness. They can be 
backed up with inexpensive repair protocols that ensure deterministic 
reliability to a distributed application. The resultant protocols are able 
to scale well in groups with thousands of participants and non-trivial node 
and packet delivery failure rates. The reliability, scalability, 
simplicity, and soft real-time traits of these protocols are a good match 
with the hardware and application requirements in emerging areas of 
peer-to-peer computing such as large-scale networks of devices.

==============================

*The talk will be followed by refreshments in Ryerson 255*

Host: Ian Foster


=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Margery Ishmael
Secretary to the Chairman, Department of Computer Science
The University of Chicago
1100 E. 58th Street, Chicago, IL. 60637-1581
tel. 773.834.8977 fax. 773.702.8487
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-