Coda File System

Re: Coda servers and WAN links?

From: Peter J. Braam <>
Date: Fri, 7 Aug 1998 09:22:56 -0400 (EDT)
Venus _could_ do this with some modifications.  In fact there are plans to
implement several such "side effects" as well as a more elaborate
mechanism using a "proxy client" to actually repeat work (ask Yui-Wah Lee
(Clement) <>) about this.

Many people have asked us about block level caching and maybe we'll do
that one day.

The primary emphasis of our project will remain on issues of stability,
performance and completion of the Windows ports first.  After that, the
world is wide open - but a lot work remains on these points.

- Peter -

On Fri, 7 Aug 1998, Bruce Janson wrote:

>     I have been thinking about this too and I was kind of interested in
>     knowing about the usage on a metered line. It seems like my coda
>     client keeps my ISDN line up all the time. Is there any reason for
>     this? Would the same thing happen to if I put a coda server here? 
>     It would be fine if it sat around not keeping the line up and then the
>     moment that it gets a request for a file, it goes back to the main
>     server for that volume and checks that the file is the same.
>     -ben
> Here, we use the Linux<->Coda kernel interface (but use
> neither Venus nor Vice).  One of our development client PCs is separated
> from its server by a (busy) 2Mbits/s WaveLAN wireless link [Hi, Robert].
> The client runs a Venus-like cache manager.  The operating system
> interface does not allow one to determine directly which blocks of a
> writable cached file have actually been modified.  Initially our cache
> manager would write back entire files on close() after a write-mode open().
> For small modifications to large files this was slow.  The latest version
> of the cache manager sacrifices local CPU cycles and cache storage space
> in an attempt to reduce this communication cost.  At first open() of a
> writable file it fetches the file from the server and then makes a second
> copy in the local cache.  After each close() the cache manager compares
> the potentially modified cache file with the saved copy and sends to
> the server only those blocks which have changed.  It then updates the
> saved copy, ready for another block-level `diff' at the next close().
> Can/Does Venus act in this mode too?
> Of course, if the cache manager could just ask the kernel which blocks
> had been changed...
> Regards,
> Bruce Janson, Basser Department of              Email:
> Computer Science, Madsen Building (F09),        Phone:  +61-2-9351-3423/4
> University of Sydney, N.S.W., 2006, AUSTRALIA   Fax:    +61-2-9351-3838
Received on 1998-08-07 09:26:33