[561] in testers
Re: cache size
daemon@ATHENA.MIT.EDU (daemon@ATHENA.MIT.EDU)
Sun Dec 24 21:41:26 1989
From: probe@ATHENA.MIT.EDU
Date: Sun, 24 Dec 89 21:41:08 -0500
To: jfc@ATHENA.MIT.EDU
Cc: epeisach@ATHENA.MIT.EDU, testers@ATHENA.MIT.EDU
In-Reply-To: John Carr's message of Fri, 22 Dec 89 15:44:04 EST,
Reply-To: Richard Basch <probe@ATHENA.MIT.EDU>
Ezra is correct in his statement that AFS is not yet a supported Athena
service. Name an Athena supported locker that resides in AFS, for
instance (sipb and watchmaker are student supported and know that AFS is
not yet supported).
In fact, we are about to undergo a heavy six-month stress test of AFS.
Granted, that at first glance, it looks like the best thing, but there
have been some serious problems that will also enter into the equation.
The network usage on a cache-miss is extremely high, and the RX protocol
that AFS uses does not back-off under conditions of a congested network.
There is little design to try to first contact servers on the local
subnet, thus adding to network load. The amount of work that has gone
into keeping AFS running has been higher than should be required.
The official policy, as stated by the directors, is still "AFS is still
experimental".
-Richard
