[OpenAFS] Re: Request for testing: NATs and 1.6.6pre*
Andrew Deason
adeason@sinenomine.net
Fri, 20 Dec 2013 10:19:31 -0600
> On Thu, 19 Dec 2013 18:28:58 -0600
> Andrew Deason <adeason@sinenomine.net> wrote:
>
> > But how do you know if this is a problem for you at all? Usually the
> > most user-visible symptom is that access to AFS hangs while a client
> > is trying to write to AFS, but a lot of different things can cause
> > that.
I should have said "hangs for about a minute". Which for some users may
be indistinguishable from "forever" :)
On Fri, 20 Dec 2013 07:29:38 +0200 (EET)
"Jukka Tuominen" <jukka.tuominen@finndesign.fi> wrote:
> Even though things work nicely usability-wise (just boot and log-in
> graphically), I still think it should have a bit smoother two-way data
> transfer behind the scene. Applications like Firefox like to write
> constantly something to a homedir which happens to be on a server. This
> sometimes freezes the application momentarily, even though the amount of
> data transferred is still modest.
>
> If you think this is the kind of configuration you're interested, and
> you can provide a patch file that works on top of this, I could try to
> test it during the weekend.
That issue is probably not relevant; the hangs/freezes I'm talking about
are usually longer, for more like a minute. Though, if you find anything
in FileLog that looks like what I mentioned, then yes, the general
environment may be of interest. Let me know?
I assume by a momentary "freeze" you mean just for a second or so. That
maybe has more to do with firefox not expecting some operations like
close() to take longer than expected. If you wanted to look into that
(outside of this thread), I would try maybe running firefox under
'strace -o /some/file.dmp -tt -T', to see which system calls are taking
a long time. If you can provide that, as well as which specific times
you see a "freeze", that could be used to provide an explanation of
what's going on. (If you manage to record that, send it to openafs-bugs
or directly to me or something; probably not to the list.)
--
Andrew Deason
adeason@sinenomine.net