Zhen Xiao
AT&T Labs

Thursday, May 5, 2005
1065 Kemper Hall
3 :10-4:00 p.m.

Refreshments to follow in 1131 Kemper

Abstract

The performance of Internet content delivery systems is important due to the increasing number of applications that rely on such systems. A wide range of techniques have been proposed, implemented, and even standardized for improving the performance of content delivery on the Internet. While many of these techniques can be very effective, it remains unclear how and when they are best used. In this talk, I will present a performance study of the commercial workload collected from a large number of Web sites hosted by a major ISP and that collected from a large group of home users connected to the Internet via a well-known cable company. Our study has found that many Web sites either do not take advantage of existing content delivery techniques or unknowingly inhibit their use. For example, most Web sites do not utilize the cache-control features of the HTTP 1.1 protocol and make indiscriminate usage of cookies. In addition, we found that the majority of multimedia content is still delivered via downloading from Web servers instead of via streaming. A substantial percentage of such downloading connections are aborted before completion due to the long waiting time.

Then I will present two solutions to address these problems. The first solution is the Cassandra toolkit that combines performance and behavioral data with an extensible simulation architecture to identify content delivery problems and predict optimization benefits. I will show how a hosting service provider can use Cassandra to estimate the benefit of using a content delivery network. The second solution is the AutoStream system that can provide streaming services automatically for multimedia objects hosted on standard Web sites. Our simulation results of AutoStream demonstrated significant reduction in network traffic and client perceived latency.

Bio

Zhen Xiao received his Ph.D. from Cornell University in January 2001. He is currently a senior technical staff member of the Distributed Systems Research Department at AT&T Labs -- Research in Florham Park, NJ. His research interests include Web technologies and content delivery, multimedia and streaming, peer-to-peer systems, security and dependability, and reliable multicast. He has published about 20 papers in leading journals and conferences, including ACM Transactions on Computer Systems, USENIX, WWW, DSN, ICDCS, and Infocom. He has also served on four NSF panels and various program committees, including WWW, Infocom, and ICDCS. He is the recipient of the AT&T Labs "Research Excellence Award" (2002) for his outstanding achievement on optimizing Internet content delivery. For more information, please visit: http://www.research.att.com/~xiao/pic.