|
We have an active/active failover cluster and plan to add more databases on these clusters but the current hardware resource will possibly need to be upgraded. How can we provide performance metrics to upgrade the hardware subsystem if the current resource is okay but adding additional databases will be a problem? How do I justify this? Please help. Thank you.
(comments are locked)
|
|
If you don't already have a baseline on performance for the current systems, that's the place to start. I'd focus on memory and disk, especially understanding how tempdb works. CPU, on most systems, is never really stressed, but memory and disk almost always will be. If you're planning on adding other systems, are those systems already in production elsewhere? Then measure them as well, again, focusing on disk and memory with a special emphasis on tempdb. Or, are they under development? Then, measure them in your dev or QA area (I usually prefer getting those measures in QA). You should be able to do this from a position of knowledge and not have to make any guesses at all. Do you have any recommendation on how I could use stress level testing without affecting production? Thanks.
Jun 20 '11 at 04:58 AM
SqlAdv
@Grant Fritchey - wont benchmarking QA only be an indication to how it might be in live? That's why I used the phrase 'best guess' if QA and Prod are different h/w config then stress from system may be different? I'm guessing there isnt a QA cluster ...
Jun 20 '11 at 05:15 AM
Fatherjack ♦♦
@Fatherjack - I agree that the QA to PROD numbers are probably an extrapolated guess but I don't think the absence of a QA cluster would skew results to any degree.
Jun 20 '11 at 05:35 AM
Blackhawk-17
@fatherjack, the cluster shouldn't matter a whit. But yes, you're still, to a degree, extrapolating, assuming that prod has 16 cpu and qa is probably a 2 cpu virtual (plucking numbers from the air). But it's still a question of performing measurements and then using those. Because you can also run production in that same environment and then simply draw straight lines between the two. It's all, at least for me, working from knowledge rather than making guesses. It's mainly that word that bothers me.
Jun 20 '11 at 05:39 AM
Grant Fritchey ♦♦
@Sqladventure. Only if you put it on another machine. But you don't really need to stress test the system, you need to measure the system you have and the apps you're bringing online. None of this is perfect, flawless, absolutely 100% correct 100% of the time. You never know when a particular set of apps might have a unique conflict on a particular resource. But you can know how much of any given resource an app uses by measuring it. That's your best bet. But, if you want to stress systems, other than prod, I'd suggest taking a look at the RML utilities. I've found the playback capabilities there to be pretty slick.
Jun 20 '11 at 05:42 AM
Grant Fritchey ♦♦
(comments are locked)
|
|
You need to make a "best Cpu and Ram.
Jun 20 '11 at 03:54 AM
SqlAdv
do you have benchmark figures for the CPU + RAM usage at the moment? If not then your first steps should be to get a record of how they are being used currently
Jun 20 '11 at 03:57 AM
Fatherjack ♦♦
(comments are locked)
|
