[643] in Release_7.7_team
ACMG: AFS Problem delaying Public Release.
daemon@ATHENA.MIT.EDU (Bill Cattey)
Thu Jul 11 12:28:45 1996
Date: Thu, 11 Jul 1996 12:28:37 -0400 (EDT)
From: Bill Cattey <wdc@MIT.EDU>
To: release-team@MIT.EDU
In the process of testing, we have uncovered a new problem with AFS.
This problem has only become apparent on machines that had been up for a
very long time under heavy load. The risks are significant enough that
the Release Team has decided to delay the release.
Two failure modes have been observed:
1. Some commands run slower.
Accessing files (including system software) is slower.
2. Machine very slow.
Doing a detailed file listing (ls -l) is slow and never caches
results on the client machine.
Failure mode 1 is a significant impact because users will get slow machines.
Failure mode 2 is a VERY significant impact because, with the lack of
caching locally of certain information, the potential exists to load
down the AFS servers and hurt everyone using AFS, not just the user on a
single workstation.
So far we have only seen this problem on Solaris machines.
It is a difficult problem to reproduce, so we MAY have this problem on
SGI's and not yet know it.
The actions we are taking are:
Delay public release until a week from Tuesday AT THE EARLIEST until we
know more about the problem.
Recommend that SPARC 5's, when they come in, be installed, deployed, but
left dark until we know more.
Work with Transarc and Sister institutions to evaluate this problem:
Reproduce the fault reliably.
Determine the exact circumstances when it can happen.
Review the likelihood of this problem happening in the clusters.
Determine a fix.
If we cannot find a fix for the problem as quickly as we would like, we
have a work-around: The problem goes away if you reboot the client
workstation.
In Team Athena, four people are making this problem their Number 1
priority. (Miki, Craig, Greg, Bill)
A defect is open at Transarc with Severity 1.
The release-team will keep ACMG informed of progress.
-wdc
