[OpenAFS] vice partition sizes, and large disks

Bill Zumach zumach@bellatlantic.net
Sat, 16 Jun 2001 09:45:07 -0400

On Sat, 16 Jun 2001, Derrick J Brashear wrote:`
> On Sat, 16 Jun 2001, chas williams wrote:
> > >I would think that the ufs logging would work ok if you ran the namei
> > >server on solaris, but have never tried it.
> > 
> > yeah, i keep forgetting to try that.  the only problem really is that
> > migrating between the regular and namei interface wont be fun.  it
> > would very handy though.
> You need one spare fileserver's worth of capacity. I'm considering it for
> our next-generation fileservers.
One of my concerns when I wrote the namei interface was if it would be too
slow when creating files. When one creates a file, the directory block
containing the name and the block containing the inode both get sync'ed to
disk. Sure this is not a problem when just one client is creating files.
But how about when a lot of clients to this at once? This works for the
current OSs using the namei interface, NT and Linux, since NT has a fast
file create operation (based on timing it) and the ext2 filesystem has
a relaxed method of updating metadata.

I'm not saying don't do this, but among your other tests before converting
I'd suggest adding some tests to time this possible performance bottleneck.

Bill Zumach

have fast metadata operations