Coda File System

Re: Venus cache size

From: Gulcu Ceki <cgu_at_zurich.ibm.com>
Date: Wed, 06 Oct 1999 11:25:52 +0200
Hi,

Reading though the mailing list, it became obvious that many are
concerned about the venus cache size. There seemed to be unanimity
about the fact that one needed a larger venus cache than the file
being manipulated under /coda.

My previous mail on the subject claimed that one could open a 20MB
file with a 10MB cache.  

Here is a description of the corroborating experiment:

- A single client running Linux RH 5.2,
- Two Servers RH 6.0, 
- One of the servers, called gilgamesh, is old (486 66Mhz, 16MB, 700MB)
- The other server, called torino, is much faster (AMD K6, 128MB, 13GB)
- Client configured with 10MB of venus cache, two servers.
- Client *does not* do hoarding.

I issued the command,

> dd if=/dev/zero of=/coda/zero bs=1024 count=20000

while both servers were up. I observed that the client would contact
both servers. Many packets were exchanged, presumably 20MB of
zeroes. However, I cannot really tell if the servers were contacted in
turn or in parallel.

I issued the command,

> filcon parition -s torino -c napoli

where torino is the faster server, and napoli is the client.  This
disconnects torino and napoli. Other connections are not
affected. BTW, isn't filcon an open invitation for denial-of-service
attack?

Then, I issued the "dd" command exactly as previously.

I observed that the client talks to "gilgamesh", the older
server. Interestingly enough, the two servers did not seem to talk to
each other. However, when I did a "ls -la /coda" on the client, there
was intense activity between the two servers, presumably synchronizing.

In both cases, the file /coda/zero was there. The client did not crash
and /coda remained accessible.  There were no apparent error
conditions either.

AFAICT, strongly connected clients (and no hoarding) can manipulate
files of any size independently of cache size. Ceki
Received on 1999-10-06 05:27:03