---

Linuxcare: Interview: Talking With Matt Welsh

[ Thanks to Matthew Cunningham for this
link. ]

In terms of making a powerful but rough new operating
system accessible to the many, few have played as pivotal a role in
the history of Linux as has Matt Welsh.
In 1993 Welsh founded
the Linux Documentation Project, a volunteer organization with the
goal of producing reliable docs for the Linux operating system. He
was the original author of the Linux Installation and Getting
Started Guide and of a number of important early HOWTOs, including
the Installation and Linux XFree86 HOWTOs. In 1995 O’Reilly &
Associates published Welsh’s Running Linux, generally acknowledged
as the all-round introduction to the OS. Welsh has also been
published in Linux Journal, Dr. Dobb’s, and serves as the Senior
Editor of Linux Magazine. Welsh is currently a Ph.D. candidate in
computer science at UC Berkeley; his research focuses on Linux
clusters and scalable Internet services. Linuxcare staffers Dave
Sifry, Jim Dennis, and Matthew Cunningham caught up with Welsh last
week for a long lunch at a Thai restaurant in Berkeley. The
occasion for the interview was the publication of Running Linux,
Third Edition, now co-authored with Lar Kaufman and Kalle
Dallheimer…”

“Linuxcare: Currently you are working on the “Millennium”
project at Berkeley?”

“Matt: Actually, now I am working on two projects, although they
are both related. Millennium is about building workstation clusters
on scale so big that you spread pieces of them out across the
campus. So, the chemistry department or the astronomy department
might have their own sort of baby clusters, and the CS department
will have their huge mondo 500-processor cluster, but it is all
viewed as one big cluster because each of those mini clusters is
connected using a gigabit link. This raises some interesting issues
because communicating between nodes within a cluster and across
clusters is going to have different performance characteristics.
You thus have to write applications that automatically determine
where they are going to run based on the performance
characteristics of the networks on top of them.”


Complete Story