Original text of orals announcement
Special University Ph.D. Oral Examination
Load-Balancing and Parallelism for the Internet
Sundar Iyer
Department of Computer Science
Stanford University
10:00-11:00AM, Tuesday, February 18th, 2003
CISX Bldg, Room 101X (Cypress Auditorium)
(Refreshments served from 9:30am)
-- The talk is open to everyone. --
Abstract:
In this talk, I will describe how load-balancing and parallelism can be
used in the design of networking systems that are scalable, fault tolerant
and give performance guarantees. Today, due to the rapid growth of the
Internet, networking systems are designed in a rather ad hoc manner. They
are not easily scalable and provide little or no guarantees. As an
example, most Internet routers do not give 100% throughput, i.e. they
cannot achieve the capacity they were designed to achieve. The lack of
scalability affects future growth, while the lack of fault tolerance and
performance guarantees hampers the deployment of new applications such
as telephony, streaming multimedia and broadcast on the Internet.
The Internet infrastructure consists of routers interconnected by links.
In order to be scalable, fault tolerant and give performance guarantees,
it is necessary that each of its components have the above capabilities.
The components of concern to a packet on the Internet, can be broadly
classified into the following areas: (1) Router Architecture, which deals
with how to design the internals of a router, (2) Packet Processing, which
deals with what arithmetic and storage operations need to be performed on
a packet, (3) Packet Switching, which deals with how to efficiently move
packets within a router and (4) Internet Routing, which deals with which
link to transfer a packet over the Internet towards its final destination.
In my talk, I shall consider design issues in each of the above areas
and present solutions by using load-balancing algorithms and readily
available commodity parallel components. I will present data gathered
from the network to motivate these solutions, show via theoretical
analysis that they give performance guarantees, and touch upon their
scalability and optimality. When required, I will also present simulation
results to suggest their effectiveness. As a result, I will attempt to
convince you that load-balancing and parallelism offer a practical way of
supporting the future design requirements of the Internet, and will play
a key role in its future growth and evolution.
