[779] in Info-AFS_Redistribution
Re: cache parameter question
daemon@ATHENA.MIT.EDU (sommerfeld@apollo.hp.com)
Mon May 18 10:32:31 1992
Date: Mon, 18 May 92 09:31:28 -0400
From: sommerfeld@apollo.hp.com
To: Marc Horowitz <marc@MIT.EDU>
Cc: <ccprl@xdm001.ccc.cranfield.ac.uk>, info-afs@transarc.com
In-Reply-To: Marc Horowitz's message of Mon, 18 May 92 08:20:01 EDT
The chunk size is a function of the file system, not the RPC
protocol... the problem is that RX starts with a *large* (16 packet *
1.5K/packet) window, and thus fires off 24KB in back-to-back packets
when sending more than 24KB at a time. This takes roughly 10 seconds
to transmit over a 19.2 line; unfortunately, RX then retransmits the
*same* 16 packets 2 seconds later. Can you say "congestion collapse"?
DCE RPC does do VJ-style slow start, so it should be in better shape
than "plain RX" over slow lines.
- Bill
