10-11.30am, Friday May 22
Room 4C-501 A/B, Holmdel
Abstract
Internet sites are often unable to serve their workload, particularly during
peak activity periods. In this presentation, I will describe two systems that
improve the scalability and performance of such sites: NetDispatcher and
Narwhal.
NetDispatcher provides scalable, fault-tolerant load balancing for IP servers.
It is implemented as an IP stack extension with a management process that
controls the allocation of requests among servers according to their real-time
load and responsiveness. NetDispatcher forwards client-to-server packets
without performing any TCP/IP header translations, and does not process
server-to-client packets. Its allocation method was proven to be efficient in
live tests, supporting some of the busiest Internet sites, such as the
official Web sites for the 1996 and 1998 Olympic Games, and the Deep Blue vs.
Kasparov matches. During the Nagano Games, NetDispatcher handled over 650
million hits (including multimedia content) during 16 days, reaching a peak
volume of 103,429 hits per minute. Peak traffic levels were sustained for one
hour, yet response time remained less than one second throughout the barrage.
I will briefly describe some of the challenges of building, operating,
securing, and managing such a Web site.
Narwhal provides transparent load-balancing of requests through firewalls and
proxies, to improve the availability and performance of TCP applications. The
Narwhal prototype consists of a monitoring tool that actively evaluates the
firewalls performance, and a proxy installed at each client that collects
local performance data and forwards traffic to the firewalls.
Contact: Subrata Mazumdar at mazum@bell-labs.com