Coda File System

Re: performance and disconnect?

From: Peter J. Braam <braam_at_cs.cmu.edu>
Date: Wed, 22 Oct 1997 08:55:55 -0400 (EDT)
Bradley,

During the summer I had to demo a variety of weakly connected (=write
disconnected) and disconnected operations to ARPA.  They revealed a
variety of bugs in the disconnection code and only through carefully
timing my script, I could make things work right. We have just tracked one
of the bugs down and will fix it this week. 

weakly connected operation should be very fast.  Lily Mummert and I
designed a write back cache scheme for Coda and Lily implemented the first
bit; this relies heavily on write disconnected operation and logging and I
never seen Coda so fast.  The implementation is not complete and the
second half is my work -- hopefully some time this academic year. 

So your observations about performance are consistent with mine and reveal
a bug.  Sorry, but we are working on it. 

- Peter -

On Tue, 21 Oct 1997, Bradley C. Kuszmaul wrote:

>    Date: Tue, 21 Oct 1997 14:57:11 -0400 (EDT)
>    From: "Yui-wah LEE (Clement)" <clement_at_ux3.sp.cs.cmu.edu>
> 
>    > I tried doing
>    >  cfs disconnect
>    > but then nothing worked.  (Couldn't write anything)
> 
>    Did you have two coda users on the client, and one of them (but not
>    you) was given the status of "primaryuser" ?  This is one of the
>    possible cause for this kind of problem.  When coda goes disconnected,
>    only one user own the client modification log (CML), which is where
>    all the local change recorded.
> 
> How would I find out?   
> 
> I am the only user who has ever done clog on the client.
> I am the only user who has files in the filesystem.
> 
> -Bradley
> 
Received on 1997-10-22 09:05:26