[OpenAFS-devel] Updated redhat workaround patch

[Derrick J Brashear]

On Tue, 22 Oct 2002, chas williams wrote:

> In message <sjm3cqywolu.fsf@kikki.mit.edu>,Derek Atkins writes:
> >Oh, he merged in that crap?  Damn.  It's a read-only,
> >no-authentication, no-vlserver-lookup AFS client.  What's
> >the point?
> 
> dont worry.  they are going to write that part.  and it will
> be bug free and prefect.  it does seem like a waste to rewrite
> something that already works and just needs a little more work
> to be really first class.

I said I was done, but I lied. I suck. IMO we need:
-vcache/inode separation, with inodes containing vcache hints that must be
verified on binding. the current situation where we have effectively our
own inode pool needs to die
-SMP-fast instead of mere SMP-safe

We *may* need:
-rx performance enhancements. nobody has yet quantified that it's really
rx and not a client or server issue, but i suspect what Hartmut has
pointed at indicates a direction: SMP linux kernel on UP machine performs
better.

We should also have:
-stop using pioctl for things which should be a sysctl 
-other than "group encoding" for PAGs, but I think Linux Security Module
has to survive for this to happen, and last week that wasn't clear.