Coda File System

Re: bandwidth measure and disconnections

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Wed, 30 Jun 2004 17:22:41 -0400
On Thu, Jun 24, 2004 at 12:48:54PM -0300, Juan Carlos Schroeder wrote:
> Well, here it is what the "where" command in gdb prints after a crash when
> enabling "write-back" cache in venus (a venus hang)

If you are referring to wbstart/wbauto, it never worked right work and
I'm actually slowly removing it.

The 'wb*' functionality tried to provide connected mode semantics for
write-disconnected mode by adding special 'writeback permit' that, when
recalled by the server, forced a reintegration as soon as another client
tried to access the same volume. However for the server to see if other
clients are interested in the volume this meant that when the writeback
was granted, we needed to break the callbacks that volume on all other
clients. As a result the other clients end up spending a lot of time
trying to revalidate the local cache state, which ends up forcing
reintegration by the client with the writeback permit.

> I forgot to say I'm using Coda Venus version 6.0.2.

The crash happened in the RPC2 library, but you didn't mention which
rpc2 version you have installed. From the arguments it looks like it
tried to send a ViceGetWBPermit rpc call, but I don't see right now how
packing those arguments (only 4 integers) can lead to a crash.

Jan
Received on 2004-06-30 17:24:52