[OpenAFS-devel] datagrams really arent big enough?
Nat Lanza
magus@cs.cmu.edu
08 Nov 2000 21:46:07 -0500
Chas Williams <chas@cmf.nrl.navy.mil> writes:
> i am not reccomending 'cache bypass' for everyone. if we ever get the mods
> i certainly wouldnt ask for it to be placed into openafs. its just
> something we thought about doing. getting >1G files from afs is very very
> slow through the cache manager. possibly something like 'afs ftp'
> might be an easier solution though. i.e. a direct read from the fileserver
> at the userspace level.
Keep in mind that intelligent caching decisions like this are still an
active research topic; there are a bunch of things you can try, but
it's hard to find a good policy that doesn't screw over some access
patterns.
Also, exposing extra knobs to the user is pretty uniformly bad. Things
should just work on their own, because in practice the user will never
ever remember to tweak stuff appropriately for their workload, if they
even know how to accurately characterize their workload.
Even if you decide "well, this is only for high-performance users who
want to tweak their system", there's a temptation to punt on making
the defaults work better, 'cause you've decided that it's okay for
users to have to modify the system and possibly their applications for
best performance.
At most, any exposure of the caching system to the user should be on a
hinting basis -- maybe allow the user to say "I will read this file
sequentially and never touch the data again. Please don't evict
anything from the cache for it", but don't require the user to do this
to get sane cache behaviour, and don't bind the system's behaviour to
what the user requested.
As far as a direct read from the fileserver at userlevel, you should
look at Arla's arlad, which is able to run at userlevel without any
kernel hooks to do things like this.
--nat
--
nat lanza --------------------- research programmer, parallel data lab, cmu scs
magus@cs.cmu.edu -------------------------------- http://www.cs.cmu.edu/~magus/
there are no whole truths; all truths are half-truths -- alfred north whitehead