Horst Birthelmer horst@riback.net
Tue, 12 Oct 2004 09:24:14 +0200

On Oct 12, 2004, at 6:45 AM, vinod kumar boppana wrote:

>  =A0
> hai all,
> I am using afs on linux 8.0 over a cluster with afs client installed=20=

> on all the compute node and afs server on the file server.
> The problem is: when a job is submitted on the cluster and whose=20
> output file will be on the AFS partition , then the user unable to see=20=

> the output file until the job executes(which has the command of=20
> closing the output file).
> but the users want to see the output in between the job execution .

If the software is self made (which I assume in a computing cluster)=20
split the output files and concat or convert them at the end.
Users can look at their files when they are ready.
AFS syncs on "close" (assuming you don't use "storebehind" which you=20
surely don't ). As long as you don't close the file you can't be sure=20
it's been written to the fileserver.

That's the problem with caching. One wants to cache a few gigabytes the=20=

other one wants to have the files in sync all the time and doesn't need=20=

it ;-)