Re: Additional CPU justification

From: Jeff Chirco <backseatdba_at_gmail.com>
Date: Tue, 21 Mar 2017 14:14:44 -0700
Message-ID: <CAKsxbLrybw7nsWFBrD0q+OBWkKBOpXEZJFUpTDkUTMkX_fy_2A_at_mail.gmail.com>



Thanks for your input. For our Linux move we already bought a new server for it so you're right it will be faster CPU's which should help. I think I'll have to wait till then and reanalyze. The thing that worries me is when we do spike to 90-100% for a couple minutes or more then people start noticing a slow down. I've tried putting some resource limits in place to prevent absolute complete take over by one database. All of our databases on this server are custom development so we have access to code and most is pretty efficient but I may have to take a finer look again.

On Tue, Mar 21, 2017 at 2:03 PM, Alfredo Abate <alfredo.abate_at_gmail.com> wrote:

> Martin is on point with newer faster chips (keeping the core counts in
> check for licensing purposes).
>
> You mention going to Linux, no slam against Windows but how will the
> database perform on Linux with the same hardware?
>
> Perhaps you can reduce the spikes by spreading the workload over a broader
> time period or maybe some additional tuning of a few key problem queries
> might get your CPU utilization at an acceptable threshold. I've seen too
> many times where hardware upgrades are used as a band aid when some tuning
> would have fixed the problem (I"m not saying this case applies). You may
> have already done the tuning exercise or maybe this is a 3rd party
> application for which you cannot tune anything.
>
> For me a 40% utilization "most of the time" is not bad. You're utilizing
> what you paid for and not leaving wasted CPU cycles. With the price of
> Oracle licensing you really have to squeeze as much out of the server as
> you can without penalizing the work to be done or more importantly the user
> experience.
>
>
> Alfredo
>
> On Tue, Mar 21, 2017 at 3:27 PM, Martin Berger <martin.a.berger_at_gmail.com>
> wrote:
>
>> Jeff,
>>
>> You can argue with the wait time of a system with 2 CPUs and a load > 40%.
>> a nice graph is here:
>> https://www.researchgate.net/figure/235937645_fig8_Waiting-t
>> ime-on-queue-2-MMm-Graph-plotting-how-the-waiting-times-in-generic-units
>>
>> Beside the theoretical background you can say, (about a 2 CPU system) at
>> 40% utilization the response time is 50% higher than at minimal
>> utilization.
>> at 70% utilization the response time is 200% higher.
>> (years ago I read "Forecasting Oracle Performance" there are some
>> backgrounds about queueing available)
>>
>> From a financial perspective I'd first check if upgrading the system with
>> faster CPUs will be much cheaper. RAM is not licensed at all, so more RAM
>> might help also - depending on your requirements.
>>
>> good luck,
>> Martin
>>
>>
>>
>> 2017-03-21 21:07 GMT+01:00 Jeff Chirco <backseatdba_at_gmail.com>:
>>
>>> Hi everyone, I am working on trying to make a case that our production
>>> database server needs some additional CPU's but was hoping you might be
>>> able to give me some tips/suggestion that you've used to prove your case.
>>> We are a smaller shop and so we currently only have a 2 CPU database
>>> license and so the cost of doubling that plus all the packs and options we
>>> have is not an easy thing to pass.
>>> We currently running 11.2.0.4 on Windows but plans to move to Oracle
>>> Linux this year and possibly 12c at that time.
>>> We occasionally spike to 80-100% during the day but average around 40%
>>>
>>> Thanks for any help.
>>>
>>> Jeff
>>>
>>
>>
>>
>

--
http://www.freelists.org/webpage/oracle-l
Received on Tue Mar 21 2017 - 22:14:44 CET

Original text of this message