Samba cluster support codebase

[James Peach]

On Fri, 2006-03-24 at 06:41 +0300, Alexander Bokovoy wrote:
> James Peach пишет:
> >>> Mine works. However, there will be certain workloads where it
> >>> won't perform as well as the standalone case.
> >> Any performance statistics you could make available? For example,
> >> for standalone case with your patch, two-node cluster, three-node?
> > 
> > Sorry, I don't have any useful numbers. For my design, any workload
> > that hits the locking TDB heavily will suffer. Although if a windows
> > client holds an exclusive oplock it won't send locking requests over
> > the wire, so at the time I didn't see this as a significant drawback.
> > 
> Could you make a simple standalone (one node) with/without the patch
> comparison?

I did make this comparison, though not in enough detail to be really
meaningful. I don't have the numbers to hand.

> In our experiments this is kind of 40-50% degrade of performance for

For what sort of workload?

> this approach, so emulating locking.tdb usage in a distributed way is
> not going to bring significant benefits. There has to be better design.

Yep, there is.

-- 
James Peach | jpeach at samba.org