[nvmewin] Performance issue

Fri Nov 22 16:02:00 PST 2013

    >> "It was subsequently removed as it was determined there may be some performance hit."

For future reference, I would ask that we add a much more explicit detailed warnings if bugs of a severe nature such as a 50% reduction in performance are re-introduced deliberately.

Imo  the sentence above does not adequately describe what we ran into and what took a lot of effort to isolate - unless one had been around as an active participant on the in October of 2012, one would not have known at all what to look for.

Thanks,
Judy

From: Alex Chang [mailto:Alex.Chang at pmcs.com]
Sent: Friday, November 22, 2013 11:38 AM
To: Knoblaugh, Rick; Kwok Kong; Judy Brock-SSI; Santosh Singh; nvmewin at lists.openfabrics.org
Subject: RE: [nvmewin] Performance issue

Thank you very much, Rick.
With the details, we can discuss more in the meeting on Monday to determine what to do with it.

Alex

From: Knoblaugh, Rick [mailto:Rick.Knoblaugh at lsi.com]
Sent: Friday, November 22, 2013 11:33 AM
To: Alex Chang; Kwok Kong; Judy Brock-SSI; Santosh Singh; nvmewin at lists.openfabrics.org<mailto:nvmewin at lists.openfabrics.org>
Subject: RE: [nvmewin] Performance issue

Hi Alex,

              Yes, the call to StorPortGetUncachedExtension was included with the initial TRIM code, so that folks evaluating the TRIM patch would not immediately hit the known assert issue. I went back and took a look at the release notes I had sent out with the patch:

In October of 2012, a patch had been put in NVMeFindAdapter to get around an
assert that was occurring on checked builds of Windows 8. It was subsequently
removed as it was determined there may be some performance hit.
Now that we are specifically testing Windows 8 features, it has been put back here as a
convenience to avoid getting the assert right out the chute. This can be taken out again if so desired.

              Given the performance issue, this should probably be again removed from the code base.  Greatly appreciate you guys revisiting this topic.

Thanks,

    -Rick

From: nvmewin-bounces at lists.openfabrics.org<mailto:nvmewin-bounces at lists.openfabrics.org> [mailto:nvmewin-bounces at lists.openfabrics.org] On Behalf Of Alex Chang
Sent: Friday, November 22, 2013 9:33 AM
To: Kwok Kong; Judy Brock-SSI; Santosh Singh; nvmewin at lists.openfabrics.org<mailto:nvmewin at lists.openfabrics.org>
Subject: Re: [nvmewin] Performance issue

Hi Rick,

The line was included in your TRIM support patch last June. Could you please elaborate why you added it?

Thanks,
Alex

From: nvmewin-bounces at lists.openfabrics.org<mailto:nvmewin-bounces at lists.openfabrics.org> [mailto:nvmewin-bounces at lists.openfabrics.org] On Behalf Of Kwok Kong
Sent: Friday, November 22, 2013 8:17 AM
To: Judy Brock-SSI; Santosh Singh; nvmewin at lists.openfabrics.org<mailto:nvmewin at lists.openfabrics.org>
Subject: Re: [nvmewin] Performance issue

Let me add this to the agenda for next week's call.

-Kwok

From: Judy Brock-SSI [mailto:judy.brock at ssi.samsung.com]
Sent: Friday, November 22, 2013 12:53 AM
To: Santosh Singh; nvmewin at lists.openfabrics.org<mailto:nvmewin at lists.openfabrics.org>
Cc: Kwok Kong
Subject: RE: [nvmewin] Performance issue

Per the attached excerpts of archived nvmewin reflector exchanges, apparently this line of code was discovered to have been causing significant performance problems a year ago and was to have been removed until the issue was understood/resolved.

Somehow it is still in the current code base and causing (us at least) the same drastic performance problems that were reported by others as far back as last November.  I'm confused as to how this was resolved for others since the code is still there....

Thanks,
Judy

From: nvmewin-bounces at lists.openfabrics.org<mailto:nvmewin-bounces at lists.openfabrics.org> [mailto:nvmewin-bounces at lists.openfabrics.org] On Behalf Of Santosh Singh
Sent: Thursday, November 21, 2013 11:09 PM
To: nvmewin at lists.openfabrics.org<mailto:nvmewin at lists.openfabrics.org>
Cc: 'Kwok Kong'
Subject: [nvmewin] Performance issue

Hi Kwok/Alex,

There is a  performance issue  in our OFA driver on  Windows 2012 and same should be applicable to Windows 8.

In the NVMeFindAdapter() function , code specifically written for the  Windows 8 under the NTDDI_VERSION > NTDDI_WIN7 .

#if (NTDDI_VERSION > NTDDI_WIN7)

/* Ensure Storport allocates the DMA adapter object */

StorPortGetUncachedExtension(pAE, pPCI, 1);

#endif

After removing this code the performance on the multi core system increases 2 times  on window 2k12 ,

and removing the 1 Byte allocation does not have any side effect on the driver functionality.

As per the Storport  documentation   the StorPortGetUncachedExtension routine allocates an uncached common

buffer to be shared by the CPU and the device. Not sure but my best guess is that  buffer sharing among CPUs could

be a reason of low performance.

Can we add this in  the known issue.

Regards

Santosh

-----Original Appointment-----

From: nvmewin-bounces at lists.openfabrics.org<mailto:nvmewin-bounces at lists.openfabrics.org> [mailto:nvmewin-bounces at lists.openfabrics.org] On Behalf Of Kwok Kong

Sent: Saturday, November 16, 2013 5:33 AM

To: nvmewin at lists.openfabrics.org<mailto:nvmewin at lists.openfabrics.org>

Subject: [nvmewin] Quarterly Status meeting

When: Monday, November 25, 2013 1:00 PM-2:00 PM (UTC-08:00) Pacific Time (US & Canada).

Where: Live Meeting

When: Monday, November 25, 2013 1:00 PM-2:00 PM (UTC-08:00) Pacific Time (US & Canada).

Where: Live Meeting

Note: The GMT offset above does not reflect daylight saving time adjustments.

*~*~*~*~*~*~*~*~*~*

Agenda:

* Review 1.3 release plan

* Windows 32-bit support

* Windows 8 extended SRB support

* NVM format state machine enhancement

* End to end protection support

* Hibernation

* Known issues

* - Not handling CSTS.RDY status (from 1->0 and 0->1) properly on NVMe reset. (Judy - Samsung)

* NUMA group support in core enumeration

* Controller reset does not handle all cases

* Learning of CPU core to Vector failure handling

* Core-MSI vector queue mapping issues

* Discuss plan for 2014

* Features that will not be supported in 2013

* ===========================================

*

* NVMe 1.1 support:

* - multi-path

* - SGL

* - Get/Set feature update

*         - Autonomous power state transition

*         - Host Identifier

*         - Reservation Notification Mask

*         - Reservation Persistence

* - identify structure update

* - write zeros command

*

Next Meeting

Please let me know if you would like to add any items to the agenda.

-Kwok

-+-----+-----+-----+-----+-----+-----+-----+-----+-

Kwok Kong has invited you to attend an online meeting using Microsoft(r) Office Communications Server.

Join the meeting

Make sure the Office Live Meeting client is installed before the meeting:

* I am connecting from inside the PMC-Sierra network

* I am connecting from outside the PMC-Sierra network

AUDIO INFORMATION

To join a meeting from your phone, dial in using the following information:

        Phone:  Burnaby Ext 6026 [English, French]

        Phone:  +1 (888) 828-7722 [English, Spanish, French]

        Phone:  +1 (604) 415-6026 [English, French]

        Find a local phone number for your region

        Conference ID:  8339027

        Passcode:       Passcode is not required.

        Note: If you have an account on this corporate network, use your PIN to join. Have you set your PIN?

TROUBLESHOOTING

Unable to join the meeting?  Start Office Live Meeting and join the meeting with the following information:

        Meeting ID:     bd6164cdeb094e6492ca14408eca1dd3

        Entry Code:     8246

        Location:       meet:sip:kwok.kong at pmcs.com;gruu;opaque=app:conf:focus:id:bd6164cdeb094e6492ca14408eca1dd3%3Fconf-key=8246

If you still cannot enter the meeting, contact support:

* Inside the PMC-Sierra network

* Outside the PMC-Sierra network

NOTICE

Office Live Meeting can be used to record meetings. By participating in this meeting, you agree that your communications may be monitored or recorded at any time during the meeting.

  << File: Untitled attachment 00195.txt >>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/nvmewin/attachments/20131123/3d6cbb73/attachment.html>