Heavy Bug Encountered on NFSv4 Server

[J. Bruce Fields]

On Wed, Jun 20, 2007 at 11:01:37PM +0200, Thomas Hoppe wrote:
> it seems like this problem is gone somehow.
> Just to sum it up:
> 
> I just switched to another release of the Gentoo 2.6.20 Kernel
> and since then, the error did not reoccur.
> I don't think that the Gentoo guys mess with the NFS code
> so the only explanation for me would be that in the vanilla
> kernel ther was a fix for something NFS 4 related.

Hm.  OK, well, let us know if you see it again or figure out anything
more.

> How does an NFS 4 server handle multiple concurrent reads
> on a single file? Say computer A is reading a file on
> an NFS server. I have the feeling that during the read,
> other Computers are blocked/queued from reading that file.
> Is that assumption true?

No.  The reads should all be able to proceed concurrently.

> My concrete scenario is that I have a webserver which exports
> the web root to 6 webservers and whenever the load is high
> I see that images load rather laggy.

And its the same image file, or a small number of image files, that are
being read?  Do you suspect a problem with NFSv4 in particular?  (Have
you seen the same behavior with NFSv3?)  Is it possible the network
between the file server and its clients (the webserver) is maxed out?

--b.