[OpenAFS-devel] Re: optimizing afs caching (was Re: datagrams really arent big enough?

[Nat Lanza]

Chas Williams <chas@cmf.nrl.navy.mil> writes:

> nfs performs 'ok' with large files (even if you arent doing
> sequential access) nothing is cached locally you are limited to
> speed of the network.  since everyone tells me that bandwidth is
> cheap now, i dont see this as a big problem.  so if i could switch
> to 'nfs mode' for certain files, that might be somewhat beneficial.

It's not actually true that NFS does no local caching. It doesn't do
any disk caching on the client side, but data blocks are stuffed into
the client's buffer cache. NFS is much less aggressive about caching
file data, though, because its consistency model is weaker; AFS uses
a callback scheme that allows clients to cache data until the server
tells them it's invalid, while in the NFS world clients must
continually revalidate their cached data.

> afs acl's are fairly complicated but most people seem to be able to
> handle them with a bit of guidance.  users arent stupid, but they arent
> going to be intuitive either.

In some cases, there's a lot of guidance, and the fact that the AFS
permission model is different from the UNIX one causes a lot of
confusion. Cache tuning knobs are less of a problem here since they're
only a performance issue and not a security one, but I still think
it's important to make sure that the default cache behaviour does the
right thing for as wide a range of operations as possible. Adding a
tweakable knob is essentially admitting that you can't make this work
right.


--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