Coda File System

Re: SCM and NON-SCM problem

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Mon, 15 Dec 2003 16:04:16 -0500
On Thu, Dec 11, 2003 at 09:33:29AM +0100, Pedro P?rezs wrote:
> Thank you very much Lionix. Both servers work correctly.
> The server NON-SCM  runs also with the client (venus). The problem is 
> that after a replicaci?n of big files (500 MB or more) the client 
> crash (we can?t list de content of /coda (coda is the volume 
> replicated)).
> It?s posible that the second machine the client crashes the machine??

That sounds strange, however... What is the size of the Coda client
cache? Coda uses whole file caching, and the total cache size is a soft
limit when we're writing to a file as the userspace cache manager only
sees the final file size when the file is closed. It will then complete
the write to the server and dump the file out of the cache to bring the
used space back to below the configured cache size.

At this point you cannot access the file, because the client cannot make
enough room in the cache to refetch the 500MB+ file from the server.

If you are dealing with several very large files, you probably want to
not only increase the number of 'cacheblocks', but also specify a
smaller than usual number of 'cachefiles' in the /etc/coda/venus.conf
file. Normally cachefiles is set to cacheblocks / 24, which would be a
lot of files for a 1GB local cache. But if your average filesize is
around the 500MB, you really would only store 2 files and possibly a
couple of directories in such a cache.

Jan
Received on 2003-12-15 16:10:58