Re: async vs. sync

[Bernd Schubert <bernd-schubert@web.de>]

[-- Attachment #1: Type: text/plain, Size: 2082 bytes --]

On Monday 26 July 2004 19:21, you wrote:
> sorry i'm not copying the list... our local exchange server changed my
> outgoing To: address a couple of weeks ago, and i still haven't gotten
> the list moderator's attention to fix up my e-mail address.

No problem, I will send a full quote to the list.

>
> > Exactly, therefore I would like to enable the sync-export
> > option again.
> > However, reducing the speed to about 1/5 of async speed for
> > large files and
> > to about 1/20 for smaller files is not an option. The people already
> > complained that compiling their projects takes ages now...
> >
> > D'oh, sorry, I forgot to write that we use the tcp-mounts and
> > nfs3. So  those
> > numbers really shouldn't happen, should they?
>
> if the server is working correctly, theoretically this should not
> happen.
>
> are you using strange mount options on the client, such as "noac" or
> "sync" ?  "async" on the client is perfectly OK to use; it's the server
> side export option that is badness.

No, I think not. Here's the fstab line for /home:
hamilton:/home /home  nfs rsize=8192,wsize=8192,hard,intr,tcp  0  0

Here's something that is for me pretty interesting (since I don't completely 
understand whats causing it):

1.) The export option is changed on the server from sync to async, 
'exportfs -rav' is executed.

2.) One client had mounted with the async-mount option, all other clients 
without this option. No client remounts /home after the the export option was 
changed on the server.

3.) All but the one client now can write to the server with 5MB/s for large 
*and* small files. However the client which has mounted with the async-mount 
option can now write with full 100MBit (so 11 MB/s) to the server.

4.) Of course, remounting /home on any client even without giving the 
async-mount option, gives back full write speed to all clients now.

Is the behaviour from 3) intented to happen? Actually I thought that a remount 
on the clients is necessary to gain back full write speed.

Thanks,
	Bernd

[-- Attachment #2: signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]