[OpenAFS] Client using static-sized kernel datastructures?

[Derek Atkins]

James L Robinson <jlrobins@uncc.edu> writes:

> Derek Atkins wrote:
> 
> >Can you try increasing the -stat parameter and trying again?  Increase
> >it to something like 15000?  It sounds like you may have an
> >application with a file descriptor leak..
> 
> Just to bite a big one here -- can someone give an
> architectural overview as to why the OpenAFS client
> appears to use static-sized kernel datastructures? Should
> not something like this be dynamically allocated in this
> day and age (tweaking MAXUSERS, recompiling the
> kernel is a thing of the past, on most modern platforms,
> at least).

Eh?  It's not static -- it's set at runtime.  It's just not
_ADJUSTABLE_ without stopping and restarting the client.

> If the answer is 'this is what was inherited from Transarc,
> why don't you send us patches that are up to snuff that
> work on all supported platforms' -- thats cool.
> But for my own personal edification, if it is some other
> compelling reason, I'd love to learn something new
> today.

This is what we inherited from Transarc.  Why don't you send us
patches that are up to snuff that work on all supported platforms?

In all seriousness, this is actually on the list of "things to do" for
Linux -- make AFS use the Linux inode table instead of AFS maintaining
its own in-kernel array.  So, I'm not even convinced it needs to be a
portable solution to be accepted.  It does, however, need to be "up to
snuff". ;)

-derek

-- 
       Derek Atkins, SB '93 MIT EE, SM '95 MIT Media Laboratory
       Member, MIT Student Information Processing Board  (SIPB)
       URL: http://web.mit.edu/warlord/    PP-ASEL-IA     N1NWH
       warlord@MIT.EDU                        PGP key available