From Twisting Country Lanes to MultiLane Ethernet SuperHighways
Stuart McRobert
Department of Computing, Imperial College, London
Abstract
This paper describes a slightly different approach to
solving network capacity problems between workstations and
servers by significantly increasing the number of conventional
Ethernet interfaces on each server from just a few to typically a
dozen or more. So rather than installing a single faster network
backbone (e.g., FDDI, ATM, Fast Ethernet, etc.) to carry all the
traffic to and from the servers, coupled with some form of step
down hubs to connect to the local workstation Ethernets, our
approach bypasses the backbone completely and brings many local
Ethernets directly to each of the servers (typically Sun Sparc
Station 10s or 20s). This technique has worked very well for our
size of operation with several file and CPU servers, 50+
workstations and around 100 X-terminals, with still room for some
further expansion too.
Over the past year this approach has been very successful in
our main teaching laboratories, significantly reducing network
congestion and providing many more well connected networks to
support both existing and additional workstations and X-
terminals, yet with fewer clients per network, so easing local
network contention problems. This, coupled with enhancements to
the workstations and servers themselves, has yielded significant
performance improvements all round and made for much happier and
contented users.
Download the full text of this paper:
ASCII (30,754 bytes)
POSTSCRIPT (245,756 bytes)
PDF (56,562 bytes)
To Become a USENIX Member, please see our
Membership Information.