[reiserfs-list] Re: [OpenAFS-devel] more on the 2.2.18pre17 SMPcpu hog/etc.

[Jeffrey Hutzelman]

On Mon, 4 Dec 2000, Andi Kleen wrote:

> > Someone suggested caching dentries for all of the cache files.  While this
> > might be portable, it would result in huge numbers of dentries being kept
> > around basically all the time, which somehow doesn't strike me as being a
> > good idea.
> 
> Alternative is to keep file names around. Keeping dentries or inodes is 
> just a shortcut. With a suitable file system structure (not too big
> directories, multiple directories) even keeping file names and opening
> on demand should be no problem, because the directory cache in the kernel 
> will cache the lookup for you.

Unfortunately, the AFS cache does not meet those conditions.  The cache is
a single directory that may have tens of thousands of files in it.  This
structure, and the interfaces between the kernel-mode cache manager and
the user-mode startup code, are consistent across many platforms and
filesystems.  Making a major architectural change for one filesystem on
one platform is not likely to happen, and could lead to serious
instability.

It is also not necessary -- it seems clear that in order to support
reiserfs, the kernel will need to cache additional information about each
inode.  Such behaviour would not be unprecedented, and would not even be
all that difficult.  I also believe that such a change, if done correctly,
has a high chance of being accepted into OpenAFS. (*)

Now, is there anyone with prior cache manager experience who wants to
work on this?  Derek?

-- Jeffrey T. Hutzelman (N3NHS) <jhutz+@cmu.edu>
   Sr. Research Systems Programmer
   School of Computer Science - Research Computing Facility
   Carnegie Mellon University - Pittsburgh, PA

(*) For present and future reference...  I happen to know and occaisonally
work with one of the OpenAFS gatekeepers.  I can often guess or ask what
his reaction to something should be.  However, that doesn't mean that my
guesses and predictions about what the gatekeepers will do should be taken
as anything more than what they are -- guesses and predictions.  Sometimes
they surprise me. :-)