Re: Client deadlock?

Hi,

I'm embarrassed to say that I saw write stalls a good while back, just moving a large file into a constrained cache--but didn't have time to approach it.  Bonnie is a complex workload--it would be attractive to find the simplest workload that can reproduce...I would try just moving large object in a small-memory vm.

Regards,

Matt

----- "Benjamin Kaduk" <[email protected]> wrote:

> On Thu, 31 Mar 2011, Garrett Wollman wrote:
> 
> >
> > It turns out that even dd(1) hangs it nearly immediately:
> >
> > $ dd if=/dev/zero of=test bs=64k count=1048576
> > load: 0.04  cmd: dd 7375 [afsslp] 17.63r 0.01u 0.45s 0% 1088k
> > load: 0.03  cmd: dd 7375 [afsslp] 21.97r 0.01u 0.45s 0% 1088k
> 
> Assuming I can do math correctly, that's creating a 64G file; it looks
> 
> like bonnie++ by default will use something like a 2G test file. 
> Since 
> your cache is roughly 1.5G, I am going to go out on a limb and surmise
> 
> that write-on-close() is working okay but earlier writes due to cache
> 
> starvation are not.
> 
> Which is kind of unfortunate.
> 
> -Ben Kaduk
> _______________________________________________
> port-freebsd mailing list
> [email protected]
> https://lists.openafs.org/mailman/listinfo/port-freebsd

-- 

Matt Benjamin

The Linux Box
206 South Fifth Ave. Suite 150
Ann Arbor, MI  48104

http://linuxbox.com

tel. 734-761-4689
fax. 734-769-8938
cel. 734-216-5309