Reduce duplicate symbols, merge server_id

[Volker Lendecke]

On Tue, May 03, 2011 at 12:13:15PM +0200, Volker Lendecke wrote:
> On Tue, May 03, 2011 at 06:06:51PM +1000, Andrew Bartlett wrote:
> > What I'm confused about is how this will interact with serverid.tdb.  I
> > my proposal to use id2, it was clear that this combines with pid and vnn
> > to form the lookup key and retrieves a unique_id.  How should this code
> > work if there are multiple unique_id values for a given (pid, vnn)
> > tuple?
> 
> The question is what we want to optimize for. If we want to
> optimize for the standard model, we could change
> serverid.tdb to have a list of unique_ids hosted by that
> process, expecting that list to be short. If want to
> optimize for the single process model, we need to key off a
> separate in-process id. Given that, you're right that we
> might need a separate ID in struct server_id. The classical
> time vs space tradeoff.

Expanding further on your question on irc whether this is
really a hot code path: Yes, it is. On every read and write
request to properly implement mandatory locking we need to
validate the brl entries that are around. It does involve
syscalls, so the relative cost of increasing that structure
might be negligible, but it is in the one hot code path we
have.

Volker

-- 
SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen
phone: +49-551-370000-0, fax: +49-551-370000-9
AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen