On 10/23/06, Andrew Sullivan <[EMAIL PROTECTED]> wrote:
> On Sat, Oct 21, 2006 at 02:08:48PM -0300, Andrew And wrote:
> > I am using Slony with 4 slaves, and I look that my CPU in master has with
> > 25% - 40 % busy when slony is active (UP). When I stop Slony the CPU is less
> > than 8% busy in Master.
>
> It sounds like your replication tables need some attention. 40% busy
> suggests to me that the system is doing a lot of sorting that it
> shouldn't need to do, which tells me you may have some stats tuning
> or vacuuming needed.  But. . .

Is that strictly CPU or is it an amalgamated measurement? I'd be quite
surprised to see that level of usage just for CPU.

> > I will need to use 80 slaves in 2 months, then I am nervous with this.
>
> . . .have you tested such a scenario?  80 replicas is a _lot_.
> There is a non-zero overhead per replica.  My guess is that 80 is
> going to be a really significant burden.  I'd consider doing
> something with cascaded replicas as well, but even so, it may be more
> than is practical.

Slony does support log shipped replication. You may find that this
helps you to support such a very large number of subscribers.

> > My master is CPU Intel Pentium 4 (3.20GHz) with 1 Gb RAM.
>
> And you can be sure that that machine is nowhere near enough for 80
> replicas.  I'd have at least an 8-way system for that kind of load.

I'd go further to suggest that you might want to consider testing a
plaform other than classic Intel. FSB is likely to be a bottleneck in
this scenario.

Drew
_______________________________________________
Slony1-general mailing list
[email protected]
http://gborg.postgresql.org/mailman/listinfo/slony1-general

Reply via email to