I strongly recommend not worrying too about much about CPU. There are lots of processes that might consume CPU but will yield to other "real" work when needed.
Every time this comes up with one of my customers, I tell them to definitely monitor CPU but not to worry about it unless there's a known problem. For example, if latency spikes were reported then maybe CPU would be a nice statistic to use in nailing down the source of the problem, but by itself a CPU spike doesn't mean much. I had one particular customer once who got really obsessed with monitoring CPU and I told them they are forbidden to look at CPU statistics ever again without my express written consent.
Back in the days of spinning disk, the most important stat was disk utilization on the aggregate. As long as it was less than 50%, you should have nice low latency response times. Above 50% things would start to deteriorate.
With all-Flash or even decently sized hybrid solutions, that doesn't happen. The disks aren't the limiting factor, it's the controller. Hosts can and will bring the CPU to extremely high levels when they need it, and on the whole multiple hosts share CPU resources nicely.
For example, let's say you have a dozen database servers. Periodically they might start doing real work and drive a storage controller to the limit. That's a good thing overall. You're getting the most out of your controller and you're not limited my spinning media. If you have 3 servers all trying to push it to the limit, each will get about 33% of the available resources. Sure, you could use QoS and keep them from using 100%, but why bother? If there is a "bully" workload that is truly interfering with getting work done, you can configure QoS, but otherwise why not let all the servers work as fast as they can?
From: toasters-bounces@teaparty.net [mailto:toasters-bounces@teaparty.net] On Behalf Of tmac Sent: Friday, June 10, 2016 6:48 PM To: NGC-msayla-mesirowfinancial.com msayla@mesirowfinancial.com Cc: toasters@teaparty.net Subject: Re: High CPU
Sound like a job for Storage Quality of Service (QOS)
create a storage qos policy for each volume....create for monitor for now, no limits Assign appropriate policy to volume. It takes effect immediately.
Look at the qos statistics to see which volume(s) are busier than the others....move if you need/want to.
--tmac
On Fri, Jun 10, 2016 at 12:28 PM, Sayla, Mustafa <MSAYLA@mesirowfinancial.commailto:MSAYLA@mesirowfinancial.com> wrote: Is there a way to find out why in a CDOT cluster controller A is always between 50 and 55% CPU usage and controller B is less than 25%. Storage wise data is evenly balance but seem like some volumes are used more heavily than others and if I can identify them I would balance them to that CPU usage is even between the controllers.
Thank you
Mustafa Sayla Visit us on the Web at mesirowfinancial.comhttp://mesirowfinancial.com
This communication may contain privileged and/or confidential information. It is intended solely for the use of the addressee. If you are not the intended recipient, you are strictly prohibited from disclosing, copying, distributing or using any of this information. If you received this communication in error, please contact the sender immediately and destroy the material in its entirety, whether electronic or hard copy. Confidential, proprietary or time-sensitive communications should not be transmitted via the Internet, as there can be no assurance of actual or timely delivery, receipt and/or confidentiality. This is not an offer, or solicitation of any offer to buy or sell any security, investment or other product.
_______________________________________________ Toasters mailing list Toasters@teaparty.netmailto:Toasters@teaparty.net http://www.teaparty.net/mailman/listinfo/toasters