updating cluster linux boxes?

yasushi@cs.washington.edu
Tue, 16 Nov 1999 13:02:55 -0800

Most of the Linux boxes now run the 2.2.7 kernel because we only have
the Transarc AFS client up to that version. Recently, I tested an
alternative AFS implementation called Arla on my desktop box for more
than a month, and I conclude it's fairly stable and usable.

I'm wondering if we could upgrade all the boxes to 2.3.* and
simultaneously switch AFS to Arla en masse. Several reasons exist for
switching; (1) 2.2.7 has many known security holes, (2) 2.2.7 network
device drivers has many known bugs, (3) 2.3.* has better file system
code, (4) 2.3.* has a "magic SysRq over serial line" feature that
allows a machine to reboot via rconsole. By the way, I'll do the labor
now that David is gone.

Arla has some problems though. One is that it does whole-file caching,
so opening a big file incurs a big delay. Other is that it sometimes
sends thousands of screwy messages to syslog (but the system still
functions). On the up side, Arla does not have the show-stopper bug in
Transarc client that crashes the kernel when many directories are
"stat"ed concurrently (if you haven't encountered this problem, run
"du /afs/cs/project" or something like that), and it doesn't lock up
the kernel when a server goes down.

Let me know your opinion. I take silence to mean approval.

yasushi