[OpenAFS] You ever seen clients POUND a tiny RO replicated volume?

[Russ Allbery]

Neulinger, Nathan <nneul@umr.edu> writes:

> Our root.cell is seeing periods where some of the replicates are getting
> accessed 2000-4000 times a second. I took a browse around and vos e'd
> root.cell.readonly on a bunch of other cells, and NO ONE shows access
> patterns like ours... 

> I'm seeing fetch-status calls from individual clients spaced at 1/100
> and 1 second apart. bangbang wait bangbang wait ... 

> Ever seen anything like that? Could that have anything to do with the
> stuff fixed in the windows client where it kills fileservers?

Yes, I've seen stuff like that before.  I've usually ended up tracking it
down to either a well-intentioned student script that is doing some while
(1) equivalent in conjunction with file access, often on multiple
machines, or a piece of badly written vendor software (yes, I know that's
redundant) that doesn't understand the concept of permission denied and
goes into a tight loop constantly retrying whatever disk write it was
trying to do.

The latter problem is depressingly common; I've seen a huge variety of
commercial mathematical and statistical packages go insane if a token
expires out from under them.

-- 
Russ Allbery (rra@stanford.edu)             <http://www.eyrie.org/~eagle/>