[dpdk-ci] [dpdklab] Re: Intel performance test is failing

Brandon Lo blo at iol.unh.edu
Thu May 14 18:12:56 CEST 2020


Hi Lihong,

I've applied the same fix to the Intel 10G machine.
It should be pulling the correct baseline values from now on.

Thanks,
Brandon

On Tue, May 12, 2020 at 1:41 AM Ma, LihongX <lihongx.ma at intel.com> wrote:

> Hi, Brandon
>
>
>
> Thanks for your help to check the issues of baseline, and I find the FVL
> env is ok, the expected value has been changed to the value we wanted.
>
> But the baseline of NNT has changed, can you help to check it ?
>
>
>
> Thanks
>
> Regards,
>
> Ma,lihong
>
>
>
> *From:* Ma, LihongX
> *Sent:* Friday, May 8, 2020 4:02 PM
> *To:* 'Brandon Lo' <blo at iol.unh.edu>
> *Cc:* Chen, Zhaoyan <zhaoyan.chen at intel.com>; David Marchand <
> david.marchand at redhat.com>; dpdklab at iol.unh.edu; Lincoln Lavoie <
> lylavoie at iol.unh.edu>; Thomas Monjalon <thomas at monjalon.net>; ci at dpdk.org;
> Tu, Lijuan <lijuan.tu at intel.com>; Xu, Qian Q <qian.q.xu at intel.com>;
> Zhang, XuemingX <xuemingx.zhang at intel.com>; O'Driscoll, Tim <
> tim.odriscoll at intel.com>; Lin, Xueqin <xueqin.lin at intel.com>
> *Subject:* RE: [dpdklab] Re: [dpdk-ci] Intel performance test is failing
>
>
>
> Thanks Brandon, wait your reply.
>
>
>
> Regards,
>
> Ma,lihong
>
>
>
> *From:* Brandon Lo [mailto:blo at iol.unh.edu <blo at iol.unh.edu>]
> *Sent:* Wednesday, May 6, 2020 9:05 PM
> *To:* Ma, LihongX <lihongx.ma at intel.com>
> *Cc:* Chen, Zhaoyan <zhaoyan.chen at intel.com>; David Marchand <
> david.marchand at redhat.com>; dpdklab at iol.unh.edu; Lincoln Lavoie <
> lylavoie at iol.unh.edu>; Thomas Monjalon <thomas at monjalon.net>; ci at dpdk.org;
> Tu, Lijuan <lijuan.tu at intel.com>; Xu, Qian Q <qian.q.xu at intel.com>;
> Zhang, XuemingX <xuemingx.zhang at intel.com>; O'Driscoll, Tim <
> tim.odriscoll at intel.com>; Lin, Xueqin <xueqin.lin at intel.com>
> *Subject:* Re: [dpdklab] Re: [dpdk-ci] Intel performance test is failing
>
>
>
> Hi Lihong,
>
>
>
> Just a further update: we have noticed that there is another internal
> script that is used to calculate baselines that will pull a newer baseline
> if it is found.
>
> We are looking to solve the issues that we are having with baselines and
> will get back to you.
>
>
>
> Thanks for your patience,
>
> Brandon
>
>
>
> On Thu, Apr 30, 2020 at 10:19 AM Brandon Lo <blo at iol.unh.edu> wrote:
>
> Hi Lihong,
>
>
>
> The expected value was reset by one of our internal scripts.
>
> I believe that I have resolved this issue for the future by ensuring that
> the baseline that you sent me will not be overwritten automatically.
>
>
>
> I will continue to monitor this expected throughput in case of any issues.
>
>
>
> Thanks for your patience,
>
> Brandon
>
>
>
> On Wed, Apr 29, 2020 at 1:30 AM Ma, LihongX <lihongx.ma at intel.com> wrote:
>
> Hi, Brandon
>
>
>
> I checked the new result of FVL, find the expected value also not changed.
>
> From the log find the expected value also is:
>
>
>
> Can you help to double check it ? Is there any different between FVL and
> NNT ?
>
>
>
> Regards,
>
> Ma,lihong
>
>
>
> *From:* Brandon Lo [mailto:blo at iol.unh.edu]
> *Sent:* Wednesday, April 29, 2020 12:52 AM
> *To:* Ma, LihongX <lihongx.ma at intel.com>
> *Cc:* Chen, Zhaoyan <zhaoyan.chen at intel.com>; David Marchand <
> david.marchand at redhat.com>; dpdklab at iol.unh.edu; Lincoln Lavoie <
> lylavoie at iol.unh.edu>; Thomas Monjalon <thomas at monjalon.net>; ci at dpdk.org;
> Tu, Lijuan <lijuan.tu at intel.com>; Xu, Qian Q <qian.q.xu at intel.com>;
> Zhang, XuemingX <xuemingx.zhang at intel.com>; O'Driscoll, Tim <
> tim.odriscoll at intel.com>; Lin, Xueqin <xueqin.lin at intel.com>
> *Subject:* Re: [dpdklab] Re: [dpdk-ci] Intel performance test is failing
>
>
>
> Hi Lihong,
>
>
>
> Sorry about that, I have reset the baseline to the values you sent in the
> previous email.
>
> I'll look to rerun tests that have failed due to the incorrect baseline.
>
>
>
> Thanks for letting me know,
>
> Brandon
>
>
>
> On Mon, Apr 27, 2020 at 11:39 PM Ma, LihongX <lihongx.ma at intel.com> wrote:
>
> Hi, Brandon
>
> I find the baseline of NNT have changed as expected, but FVL still same as
> before.
>
> Can you help to check it and change the baseline as expected ?
>
>
>
> Regards,
>
> Ma,lihong
>
>
>
> *From:* Brandon Lo [mailto:blo at iol.unh.edu]
> *Sent:* Friday, April 3, 2020 2:39 AM
> *To:* Ma, LihongX <lihongx.ma at intel.com>
> *Cc:* Chen, Zhaoyan <zhaoyan.chen at intel.com>; David Marchand <
> david.marchand at redhat.com>; dpdklab at iol.unh.edu; Lincoln Lavoie <
> lylavoie at iol.unh.edu>; Thomas Monjalon <thomas at monjalon.net>; ci at dpdk.org;
> Tu, Lijuan <lijuan.tu at intel.com>; Xu, Qian Q <qian.q.xu at intel.com>;
> Zhang, XuemingX <xuemingx.zhang at intel.com>; O'Driscoll, Tim <
> tim.odriscoll at intel.com>
> *Subject:* Re: [dpdklab] Re: [dpdk-ci] Intel performance test is failing
>
>
>
> Hi Lihong,
>
>
>
> I have changed the baselines to reflect the new expected values.
>
> The performance tests should work as expected and pass.
>
>
>
> We will email again in the future if we come across any problems.
>
> Feel free to email us as well if you would like to make any other changes.
>
>
>
> Thank you for all your help
>
>
>
> On Wed, Apr 1, 2020 at 2:00 AM Ma, LihongX <lihongx.ma at intel.com> wrote:
>
> Hi, Brandon
>
> Thanks for you recommends, I have done the changes.
>
> As the throughput  value of nic_single_core is proportional to the cpu
> frequency.
>
> I recommend you can change the baseline according to our report system.
>
>
>
> On the our 2.50GHz system, the baseline value as below:
>
> NNT:
>
> *pkt_size*
>
> *trd/rxd*
>
> *expected_value*
>
> 64
>
> 512
>
> 52.562
>
> 64
>
> 2048
>
> 41.439
>
>
>
> FVL:
>
> *pkt_size*
>
> *trd/rxd*
>
> *expected_value*
>
> 64
>
> 512
>
> 59.608
>
> 64
>
> 2048
>
> 47.73
>
>
>
> For the testbed in UNH, it’s a 2.1Ghz CPU server, so the expected number
> should be
>
> NNT:
>
> *pkt_size*
>
> *trd/rxd*
>
> *expected_value*
>
> 64
>
> 512
>
> 52.562 / 2.5 * 2.1=44.152
>
> 64
>
> 2048
>
> 41.439 / 2.5 * 2.1=34.809
>
>
>
> FVL:
>
> *pkt_size*
>
> *trd/rxd*
>
> *expected_value*
>
> 64
>
> 512
>
> 59.608 / 2.5 * 2.1=50.071
>
> 64
>
> 2048
>
> 47.73 / 2.5 * 2.1=40.093
>
>
>
>
>
>
>
> Regards,
>
> Ma,lihong
>
>
>
> *From:* Brandon Lo [mailto:blo at iol.unh.edu]
> *Sent:* Tuesday, March 31, 2020 9:42 PM
> *To:* Chen, Zhaoyan <zhaoyan.chen at intel.com>
> *Cc:* David Marchand <david.marchand at redhat.com>; dpdklab at iol.unh.edu;
> Lincoln Lavoie <lylavoie at iol.unh.edu>; Thomas Monjalon <
> thomas at monjalon.net>; ci at dpdk.org; Tu, Lijuan <lijuan.tu at intel.com>; Xu,
> Qian Q <qian.q.xu at intel.com>; Ma, LihongX <lihongx.ma at intel.com>; Zhang,
> XuemingX <xuemingx.zhang at intel.com>; O'Driscoll, Tim <
> tim.odriscoll at intel.com>
> *Subject:* Re: [dpdklab] Re: [dpdk-ci] Intel performance test is failing
>
>
>
> Hi Zhaoyan,
>
>
>
> To make changes to either Intel machine, please reboot using the command
> "reboot_to_rw" as root to reboot the machine into read/write mode.
>
> This command will also disable any testing on the machine.
>
>
>
> To re-enable the machine, please run "reboot_to_ro" as root, and it will
> save all of the changes that you've made and re-enable testing on the
> machine.
>
> I recommend rebooting using either "reboot_to_rw" or "reboot_to_ro"
> instead of the normal "reboot" while you're making changes.
>
>
>
> After you're done, please let me know. I'll have to manually run a test
> and update the baseline using our internal CI.
>
>
>
> Thank you
>
>
>
> On Mon, Mar 30, 2020 at 12:43 AM Chen, Zhaoyan <zhaoyan.chen at intel.com>
> wrote:
>
> Hi, Brandon,
>
>
>
> Please let me know how to make change to this reset machine.
> (ip/access...) and disable it.
>
>
>
> After that please help to change the baseline.
>
>
>
>
>
> *Regards,*
>
> *Zhaoyan Chen*
>
>
>
> *From:* Brandon Lo <blo at iol.unh.edu>
> *Sent:* Thursday, March 26, 2020 11:39 PM
> *To:* Chen, Zhaoyan <zhaoyan.chen at intel.com>
> *Cc:* David Marchand <david.marchand at redhat.com>; dpdklab at iol.unh.edu;
> Lincoln Lavoie <lylavoie at iol.unh.edu>; Thomas Monjalon <
> thomas at monjalon.net>; ci at dpdk.org; Tu, Lijuan <lijuan.tu at intel.com>; Xu,
> Qian Q <qian.q.xu at intel.com>; Ma, LihongX <lihongx.ma at intel.com>; Zhang,
> XuemingX <xuemingx.zhang at intel.com>; O'Driscoll, Tim <
> tim.odriscoll at intel.com>
> *Subject:* Re: [dpdklab] Re: [dpdk-ci] Intel performance test is failing
>
>
>
> Hi Zhaoyan,
>
>
>
> Currently, we have a system in place that resets any changes made while
> testing is enabled for a machine.
>
> If you would like, I can disable testing and allow you to make permanent
> changes.
>
>
>
> I can also reset the baseline of Intel 10G test performance once you make
> these changes.
>
> Please let me know if you would like to make permanent changes on the
> Intel 10G so I can disable it for you.
>
>
>
> Thanks
>
>
>
> On Wed, Mar 25, 2020 at 12:59 AM Chen, Zhaoyan <zhaoyan.chen at intel.com>
> wrote:
>
> Thanks. Brandon.
>
>
>
> That’s good. We have made changed on 10G testbed.
>
>
>
> I monitored the several execution results; I found the results of 10G
> always has -0.9%~-1.x% gap against expected number. So it could lead to see
> sometime failures..+-1% I suggest adjusting the expected number. I don’t
> know where the expected number is from? as I know it a dynamic number?
> depends on baseline.. Please help to clarify, thanks.
>
>
>
>
>
> Thanks.
>
>
>
> *Regards,*
>
> *Zhaoyan Chen*
>
>
>
> *From:* Brandon Lo <blo at iol.unh.edu>
> *Sent:* Tuesday, March 24, 2020 9:31 PM
> *To:* Chen, Zhaoyan <zhaoyan.chen at intel.com>
> *Cc:* David Marchand <david.marchand at redhat.com>; dpdklab at iol.unh.edu;
> Lincoln Lavoie <lylavoie at iol.unh.edu>; Thomas Monjalon <
> thomas at monjalon.net>; ci at dpdk.org; Tu, Lijuan <lijuan.tu at intel.com>; Xu,
> Qian Q <qian.q.xu at intel.com>; Ma, LihongX <lihongx.ma at intel.com>; Zhang,
> XuemingX <xuemingx.zhang at intel.com>
> *Subject:* Re: [dpdklab] Re: [dpdk-ci] Intel performance test is failing
>
>
>
> Hi Zhaoyan,
>
>
>
> I have enabled the 10G Intel machine for testing.
>
> If you would like to make any more changes, please let me know so I can
> perform the necessary steps to prepare the machine for changes.
>
> Please feel free to let me know if you need anything.
>
>
>
> Thank you
>
>
>
> On Sun, Mar 22, 2020 at 9:58 PM Chen, Zhaoyan <zhaoyan.chen at intel.com>
> wrote:
>
> Hi, Brandon,
>
>
>
> For 10G, please enable it. our code is at original path
> */opt/test-harness/dts.*
>
>
>
> For 40G, please keep running. and see if any issue. But, anyway, we have
> modified the DTS code at /opt/test-harness/dts-new-suite. If we met same
> problem, then use this new DTS instead.
>
>
>
> Thanks a lot
>
>
>
> *Regards,*
>
> *Zhaoyan Chen*
>
>
>
> *From:* Brandon Lo <blo at iol.unh.edu>
> *Sent:* Saturday, March 21, 2020 1:49 AM
> *To:* Chen, Zhaoyan <zhaoyan.chen at intel.com>
> *Cc:* David Marchand <david.marchand at redhat.com>; dpdklab at iol.unh.edu;
> Lincoln Lavoie <lylavoie at iol.unh.edu>; Thomas Monjalon <
> thomas at monjalon.net>; ci at dpdk.org; Tu, Lijuan <lijuan.tu at intel.com>; Xu,
> Qian Q <qian.q.xu at intel.com>; Ma, LihongX <lihongx.ma at intel.com>; Zhang,
> XuemingX <xuemingx.zhang at intel.com>
> *Subject:* Re: [dpdklab] Re: [dpdk-ci] Intel performance test is failing
>
>
>
> Hi Zhaoyan,
>
>
>
> Currently, the 40G machine is stable enough to be put on production
> dashboard to run tests which may cause Trex to be killed.
>
> Should I disable the 40G Intel machine for you to make changes?
>
>
>
> Also, just for confirmation: on the 10G machine, is the folder that you
> are using for the testing located in */opt/test-harness/dts-2020-3-4, o*r
> are you still using the one in the standard */opt/test-harness/dts*
> folder?
>
>
>
> If everything is ok, I will enable the 10G machine for production testing.
>
>
>
> Thank you very much
>
>
>
> On Thu, Mar 19, 2020 at 9:36 PM Chen, Zhaoyan <zhaoyan.chen at intel.com>
> wrote:
>
> Brandon,
>
>
>
> We worked out a workaround on Intel testbeds. NNT(10G) and FVL(40G). Could
> you please help to recover them?
>
>
>
> But, for FVL(40G) testbed,  we met some problems, could you please help to
> check before recover it
>
>    - Sometime 1G hugepage will be changed to 2Mhugepage
>    automatically...we have to restart the system
>    - When we debugging on the testbed, found that Trex was killed by some
>    one(app)..
>
> Please help to check if any other program running on the testbed.
>
>
>
> Thanks a lot.
>
>
>
>
>
>
>
> *Regards,*
>
> *Zhaoyan Chen*
>
>
>
> *From:* Chen, Zhaoyan <zhaoyan.chen at intel.com>
> *Sent:* Wednesday, March 18, 2020 9:04 PM
> *To:* Brandon Lo <blo at iol.unh.edu>
> *Cc:* David Marchand <david.marchand at redhat.com>; dpdklab at iol.unh.edu;
> Lincoln Lavoie <lylavoie at iol.unh.edu>; Thomas Monjalon <
> thomas at monjalon.net>; ci at dpdk.org; Tu, Lijuan <lijuan.tu at intel.com>; Xu,
> Qian Q <qian.q.xu at intel.com>; Chen, Zhaoyan <zhaoyan.chen at intel.com>
> *Subject:* RE: [dpdklab] Re: [dpdk-ci] Intel performance test is failing
>
>
>
> Brandon, we almost made a workaround.
>
>
>
> Maybe tomorrow, you could recover Intel’s testbed. I will let you know
> soon.
>
>
>
>
>
>
>
> *Regards,*
>
> *Zhaoyan Chen*
>
>
>
> *From:* Brandon Lo <blo at iol.unh.edu>
> *Sent:* Wednesday, March 18, 2020 3:34 AM
> *To:* Chen, Zhaoyan <zhaoyan.chen at intel.com>
> *Cc:* David Marchand <david.marchand at redhat.com>; dpdklab at iol.unh.edu;
> Lincoln Lavoie <lylavoie at iol.unh.edu>; Thomas Monjalon <
> thomas at monjalon.net>; ci at dpdk.org; Tu, Lijuan <lijuan.tu at intel.com>; Xu,
> Qian Q <qian.q.xu at intel.com>
> *Subject:* Re: [dpdklab] Re: [dpdk-ci] Intel performance test is failing
>
>
>
> Hi Zhaoyan,
>
>
>
> Have you finished making changes on the Intel machine?
>
> I will turn on the machine on March 3rd for testing if you do not have any
> issues with it.
>
> Please let me know if you need anything else.
>
>
>
> Thanks
>
>
>
> On Tue, Mar 10, 2020 at 10:13 PM Chen, Zhaoyan <zhaoyan.chen at intel.com>
> wrote:
>
> Hi, Brandon,
>
>
>
> Yes, it’s a wired issue. And it also mixed our DTS upgrading and Trex
> upgrading.
>
> So we are reviewing our DTS script, different Trex version, and CI calling
> procedure.
>
>
>
> Anyway, we are focusing on this task recently, any update will let you
> know.
>
>
>
> Thanks.
>
>
>
> *Regards,*
>
> *Zhaoyan Chen*
>
>
>
> *From:* Brandon Lo <blo at iol.unh.edu>
> *Sent:* Tuesday, March 10, 2020 10:46 PM
> *To:* David Marchand <david.marchand at redhat.com>
> *Cc:* Chen, Zhaoyan <zhaoyan.chen at intel.com>; dpdklab at iol.unh.edu;
> Lincoln Lavoie <lylavoie at iol.unh.edu>; Thomas Monjalon <
> thomas at monjalon.net>; ci at dpdk.org; Tu, Lijuan <lijuan.tu at intel.com>; Xu,
> Qian Q <qian.q.xu at intel.com>
> *Subject:* Re: [dpdklab] Re: [dpdk-ci] Intel performance test is failing
>
>
>
> Hi Zhaoyan,
>
>
>
> How is the current status of the Intel 82599ES?
>
> Were there any configuration changes made to fix performance issues?
>
>
>
> Thanks
>
>
>
> On Tue, Mar 10, 2020 at 9:11 AM Brandon Lo <blo at iol.unh.edu> wrote:
>
> Hi David,
>
>
>
> This was just a weird issue with the packet generator not cleaning itself
> after a test fast enough before another test.
>
> I'll rerun the tests that were affected and keep an eye out to see if it's
> stable enough to be put back online.
>
>
>
> Thanks
>
>
>
> On Tue, Mar 10, 2020 at 5:33 AM David Marchand <david.marchand at redhat.com>
> wrote:
>
> On Tue, Mar 3, 2020 at 3:14 PM Brandon Lo <blo at iol.unh.edu> wrote:
> >
> > Hi David and Zhaoyan,
> >
> >
> > Yes, those results are related to the Intel machine; I have disabled
> testing for the Intel testbed.
> >
> > The 82599ES machine is now available for ssh and modifications.
>
> Any news about this?
>
> I received a failure on a patch of mine (changing macros in a ARM header).
> https://lab.dpdk.org/results/dashboard/patchsets/9900/
>
> But this time, it is with the 40G Intel nic test.
>
> --
> David Marchand
>
>
>
>
> --
>
> Brandon Lo
>
> UNH InterOperability Laboratory
>
> 21 Madbury Rd, Suite 100, Durham, NH 03824
>
> blo at iol.unh.edu
>
> www.iol.unh.edu
>
>
>
>
> --
>
> Brandon Lo
>
> UNH InterOperability Laboratory
>
> 21 Madbury Rd, Suite 100, Durham, NH 03824
>
> blo at iol.unh.edu
>
> www.iol.unh.edu
>
>
>
>
> --
>
> Brandon Lo
>
> UNH InterOperability Laboratory
>
> 21 Madbury Rd, Suite 100, Durham, NH 03824
>
> blo at iol.unh.edu
>
> www.iol.unh.edu
>
>
>
>
> --
>
> Brandon Lo
>
> UNH InterOperability Laboratory
>
> 21 Madbury Rd, Suite 100, Durham, NH 03824
>
> blo at iol.unh.edu
>
> www.iol.unh.edu
>
>
>
>
> --
>
> Brandon Lo
>
> UNH InterOperability Laboratory
>
> 21 Madbury Rd, Suite 100, Durham, NH 03824
>
> blo at iol.unh.edu
>
> www.iol.unh.edu
>
>
>
>
> --
>
> Brandon Lo
>
> UNH InterOperability Laboratory
>
> 21 Madbury Rd, Suite 100, Durham, NH 03824
>
> blo at iol.unh.edu
>
> www.iol.unh.edu
>
>
>
>
> --
>
> Brandon Lo
>
> UNH InterOperability Laboratory
>
> 21 Madbury Rd, Suite 100, Durham, NH 03824
>
> blo at iol.unh.edu
>
> www.iol.unh.edu
>
>
>
>
> --
>
> Brandon Lo
>
> UNH InterOperability Laboratory
>
> 21 Madbury Rd, Suite 100, Durham, NH 03824
>
> blo at iol.unh.edu
>
> www.iol.unh.edu
>
>
>
>
> --
>
> Brandon Lo
>
> UNH InterOperability Laboratory
>
> 21 Madbury Rd, Suite 100, Durham, NH 03824
>
> blo at iol.unh.edu
>
> www.iol.unh.edu
>
>
>
>
> --
>
> Brandon Lo
>
> UNH InterOperability Laboratory
>
> 21 Madbury Rd, Suite 100, Durham, NH 03824
>
> blo at iol.unh.edu
>
> www.iol.unh.edu
>
>
>
>
> --
>
> Brandon Lo
>
> UNH InterOperability Laboratory
>
> 21 Madbury Rd, Suite 100, Durham, NH 03824
>
> blo at iol.unh.edu
>
> www.iol.unh.edu
>


-- 

Brandon Lo

UNH InterOperability Laboratory

21 Madbury Rd, Suite 100, Durham, NH 03824

blo at iol.unh.edu

www.iol.unh.edu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mails.dpdk.org/archives/ci/attachments/20200514/88dcacd9/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.png
Type: image/png
Size: 4767 bytes
Desc: not available
URL: <http://mails.dpdk.org/archives/ci/attachments/20200514/88dcacd9/attachment-0001.png>


More information about the ci mailing list