[OpenAFS] Re: why afs backup is so poorly supported

[Jeffrey Hutzelman]

On Tuesday, October 10, 2006 09:03:40 AM -0400 chas williams - CONTRACTOR 
<chas@cmf.nrl.navy.mil> wrote:

> In message <x3vemtjc2g.fsf@nowhere.com>,Adam Megacz writes:
>> Sorry for the confusion.  It's not the way the layout is done, but
>> that it is done at all.
>
> as i said before, it likely the layout was done to reduce directory
> contention.  if you have a file buried eight levels deep the fileserver
> doesnt need to traverse eight directories to find your file.  further,
> afs doesnt know what underlying limitations your local filesystem is
> going to have with respect to number of files in a directory.
>
> usenet servers have understood this problem for sometime and the
> afs solution seems similar.

In fact, AFS came to the right solution (punt the whole name thing and 
access files by inode number) well before usenet servers did.  Perhaps 
that's because the people designing AFS were hacking on the kernel anyway, 
and adding icreate/iopen/irdwr/iincdec was trivial compared to the cache 
manager.

On the other hand, usenet servers have soundly beaten us to taking the next 
logical step, which is to punt entirely on using a general-purpose 
filesystem and instead use a purpose-built data store.

-- Jeff