[31] in Info-AFS_Redistribution
Re: AFS over SLIP
daemon@ATHENA.MIT.EDU (daemon@ATHENA.MIT.EDU)
Fri Dec 21 16:13:02 1990
Date: Fri, 21 Dec 90 15:52:50 -0500
To: Bill Cattey <wdc@MIT.EDU>
Cc: Craig_Everhart@transarc.com
Cc: info-afs@transarc.com
In-Reply-To: Bill Cattey's message of Fri, 21 Dec 1990 14:21:28 -0500 (EST),
From: Richard Basch <probe@MIT.EDU>
Date: Fri, 21 Dec 1990 14:21:28 -0500 (EST)
From: Bill Cattey <wdc@ATHENA.MIT.EDU>
References: <8bQXenQ91E5zApZwV5@rchland.ibm.com>,
<8bQYdtT0BwwOAl7Y1Z@transarc.com>
I heard stories about an MIT living group attempting to use AFS over a
9600 baud line.
Often the result was the AFS client kernel panic'ing due to running out
of kernel stack space. I think that problem was fixed by increasing the
size of the kernel stack. We now work around the problem through the
gross and horrible expedient of using an AFS to NFS translator.
I'm interested in seeing rx improve in the 9.6 case myself.
-wdc
The panic no longer occurs with the AFS 3.0 product code... instead, the
machine hangs, constantly retransmitting packets, since the responses
don't come before the RX timeout. A 14.4k link is just too slow for RX.
The first thought that comes to mind is to add slow-start/decay code as
Van Jacobsen did with TCP...
-Richard
