RE: Re: [ntdev] Re: [ntdev] SDV encountered errors when building th

I don’t know what driver the NVMe vendor was using, but the OFA driver in its original form, could be a huge CPU hog. I saw situations on a 64-core processor the over 50 cores were spinning on locks, because the locking was so poorly designed. This could be the problem you were seeing, some simple cleanup of the locking model significantly reduced the CPU utilization.

Don Burn
Windows Driver Consulting
Website: http://www.windrvr.com

-----Original Message-----
From: xxxxx@lists.osr.com [mailto:xxxxx@lists.osr.com] On Behalf Of Marion Bond
Sent: Thursday, April 09, 2015 6:01 PM
To: Windows System Software Devs Interest List
Subject: [ntdev] Re: [ntdev] Re: [ntdev] SDV encountered errors when building th

I don’t have the test results at hand, but the problems weren’t so much that the host CPU usage was too high for a given workload, but that in CPU starvation conditions, the IOOPs dropped off dramatically. This may be tied to the specific hardware I was using (and the corresponding vendor drivers), but that was my experience

In my environment, the workload is alternately CPU and IO intensive and there is no good way to predict when each condition will occur. As with any system tied to market data, the only thing you can say for sure is that the work comes in bursts. On the application level there is logic to coalesce and drop work beyond the physical capacity so that latency is not sacrificed too much, but certain parts cannot be dropped.

Sent from Surface Pro

From: xxxxx@osr.com
Sent: ‎Thursday‎, ‎April‎ ‎09‎, ‎2015 ‎9‎:‎00‎ ‎AM
To: Windows System Software Devs Interest List mailto:xxxxx



Really!? You’re seeing LESS net host CPU usage from storing your data on a SAN with SSDs than you are with NVMe disks, for the same I/O performance?

If so, that’s fascinating and useful data. Not many of us can mount that size test system, and being much more than casually involved with NVMe (cough), that would be an interesting data point.

Then again… maybe the way our database tables are constructed, once things are set there’s relatively little I/O? (note use of word relative)… hot data stays hot, for example…

There should be an interesting application for storage tiering here.

Sorry… but I’m sure we all like to play architect in somebody else’s sandbox occasionally. It’s the dilettante in us.

Peter
OSR
@OSRDrivers


NTDEV is sponsored by OSR

Visit the list at: http://www.osronline.com/showlists.cfm?list=ntdev

OSR is HIRING!! See http://www.osr.com/careers

For our schedule of WDF, WDM, debugging and other seminars visit:
http://www.osr.com/seminars

To unsubscribe, visit the List Server section of OSR Online at http://www.osronline.com/page.cfm?name=ListServer


NTDEV is sponsored by OSR

Visit the list at: http://www.osronline.com/showlists.cfm?list=ntdev

OSR is HIRING!! See http://www.osr.com/careers

For our schedule of WDF, WDM, debugging and other seminars visit:
http://www.osr.com/seminars

To unsubscribe, visit the List Server section of OSR Online at http://www.osronline.com/page.cfm?name=ListServer</mailto:xxxxx>