[6746] in Athena Bugs
Re: Who does this go to?
daemon@ATHENA.MIT.EDU (daemon@ATHENA.MIT.EDU)
Sun Dec 30 03:02:34 1990
Date: Sun, 30 Dec 90 03:01:42 -0500
To: marc@MIT.EDU
Cc: athena-ws@MIT.EDU, testers@MIT.EDU, bugs@MIT.EDU, cfields@MIT.EDU
In-Reply-To: Marc Horowitz's message of Sat, 29 Dec 90 23:04:34 EST,
From: Richard Basch <probe@MIT.EDU>
In order to run AFS 3.1, you must compile a new kernel, and get a new
"afsd". This command will not work with AFS 3.0 (which is what is in
the 7.2 release). I may or may not already have a kernel, depending on
what type of machine it is intended. However, building further kernels
will be a problem, as I am in the midst of doing a major revamping on my
kernel sources to try integrating NFS 4.0, and merge more of the
networking code between the RT and VAX (the latter is done, the prior
isn't).
-Richard
Reply-To: marc@MIT.EDU
X-Usmail: Marc Horowitz, 3 Ames Street, Cambridge, MA 02139
X-Phone: (617)225-6458
Date: Sat, 29 Dec 90 23:04:34 EST
From: Marc Horowitz <marc@ATHENA.MIT.EDU>
The right people have probably seen this, but I'll send it around to
the athena lists anyway. Someone might want to try this for the ILG
clusters. Craig, you can give it a shot at pika. As the guy said, a
slow reliable connection beats a fast unreliable one. You should also
do a speed comparison with the AFS->NFS translator, remembering that
this way, you still get the benefits of caching.
Marc
[0034] daemon@ATHENA.MIT.EDU Info-AFS_Redistribution 12/26/90 14:06 (27 lines)
Subject: More on AFS over slip - getting closer
Reply-To: Bob Oesterlin <oester@rchland.iinus1.ibm.com>
Date: Wed, 26 Dec 1990 08:03:46 -0600 (CST)
From: Bob Oesterlin <oester@rchland.iinus1.ibm.com>
To: info-afs@transarc.com
Thanks to Peter Honeyman for pointing out that altering the chunk size
may be the trick - It is!. The default chunk size for a disk cache is
64K, way more that you can ever xfer in the time-out allowed by RX. With
AFS 3.1, you cat set this with afsd:
/usr/vice/etc/afsd -chunk 10
This sets the cunk size to 1K (11 would be 2K). This seems to allow me
to actually execute commands that are in AFS, where before all I go were
core dumps. I'll take a slow, reliable connection over a faster unstable
connection any day!
The other trick that Peter suggested was to set up the SLIP MTU to 576
(in net/if_sl.c if you're using Van's TCP header compression code for
slip). He also suggested that rx.c be modified to use a packet size of
576 if the IP address was not "local" - prehaps he can post the code
here if he's able.
In any case, I'm going to experiment some more. It sure would be helpful
if the folks at Transarc could comment on all of this!
Bob
--[0034]--
