[p4] Slow performance

[Robert Cowham]

Please note that CPU performance *does* affect things like
checkpoint/restore time. Most other operations are not affected - it's more
I/O, filesystem, RAM etc.

This has been discussed on the list in the past.

I have been with a client where some not cheap Sun machines were trialled.
With a large (>100Gb) database, the SPARC based machines took at least 2x as
long as the Opteron based machines, both running Solaris 10.

Regards
Robert

> -----Original Message-----
> From: perforce-user-bounces at perforce.com 
> [mailto:perforce-user-bounces at perforce.com] On Behalf Of Benjamin Nias
> Sent: 27 February 2007 17:07
> To: duncan.bowring at realtimeworlds.com; perforce-user at perforce.com
> Subject: Re: [p4] Slow performance
> 
> Hi Duncan,
> 
> You could try checking that P4 doesn't have a CPU affinity. 
> More likely, when was the last time you 
> checkpointed/rebalanced the database? This can have HUGE 
> effects on the performance of P4.
> 
> Ben
> 
> 
> 
> -----Original Message-----
> From: perforce-user-bounces at perforce.com
> [mailto:perforce-user-bounces at perforce.com] On Behalf Of 
> duncan.bowring at realtimeworlds.com
> Sent: 27 February 2007 13:29
> To: perforce-user at perforce.com
> Subject: [p4] Slow performance
> 
> 1 user, 1 client.
> I'm noticing that when submitting/getting or creating a 
> checkpoint, it isn't really that fast.. but when looking at 
> the hardware usage, disk IO isn't very high.. 10-60meg/sec 
> and that's only occasional. It's not using all the RAM (so 
> it's not paging), when submitting it uses 100% CPU on 1 core, 
> but any other action it uses about 3-5%.
> 
> I can't see anything bottlenecking, but have you got any idea 
> why Perforce operates like this? Is this normal, will it be 
> like this with everyone, except now I'll be able to scale to 
> lots of users without any further slow down? Network activity 
> is never really much more than 60mbit, and again, that's occasional.
> 
> Any idea why in a single operation it's not really that fast 
> (or hardware dependent)?
> 
> Thanks,
> Duncan