[dts] [dpdklab] Re: [PATCH] tests/nic_single_core_perf: modify accepted_tolerance to percentage

[Tu, Lijuan]

Hi Owen,

I ‘d like to add 2 more changes in V2:

1,  test_interval  will be renamed to throughput_stat_sample_interval for better understanding.
2,  accepted_tolerance will be a float to algin with delta in json, both format are {:.2f}


From: dts <dts-bounces at dpdk.org> On Behalf Of Tu, Lijuan
Sent: 2021年3月25日 15:29
To: Owen Hilyard <ohilyard at iol.unh.edu>
Cc: David Liu <dliu at iol.unh.edu>; Ali Alnubani <alialnu at nvidia.com>; dts at dpdk.org; dpdklab at iol.unh.edu
Subject: Re: [dts] [dpdklab] Re: [PATCH] tests/nic_single_core_perf: modify accepted_tolerance to percentage

sure, I prefer to use a float too.

From: Owen Hilyard <ohilyard at iol.unh.edu<mailto:ohilyard at iol.unh.edu>>
Sent: 2021年3月24日 1:24
To: Tu, Lijuan <lijuan.tu at intel.com<mailto:lijuan.tu at intel.com>>
Cc: David Liu <dliu at iol.unh.edu<mailto:dliu at iol.unh.edu>>; Ali Alnubani <alialnu at nvidia.com<mailto:alialnu at nvidia.com>>; dts at dpdk.org<mailto:dts at dpdk.org>; dpdklab at iol.unh.edu<mailto:dpdklab at iol.unh.edu>
Subject: Re: [dpdklab] Re: [dts] [PATCH] tests/nic_single_core_perf: modify accepted_tolerance to percentage

Hello Lijuan,

Would it be possible to have the delta be a float in the json representation? For example, a 20% delta would be encoded as "delta": .2 in json. Currently, most of the community lab's infrastructure expects to make pass/fail determinations on a numeric type, so having the delta as a float would reduce the required downtime to change everything over to percentages since we shouldn't need to patch all of our services to expect a string.


Thanks,
Owen Hilyard

On Tue, Mar 23, 2021 at 10:53 AM David Liu <dliu at iol.unh.edu<mailto:dliu at iol.unh.edu>> wrote:
Thank you for Ali and Lijuan.

We had tested the patch on the Arm machines setup. Other than the value issues, the test is running as expected.

On Tue, Mar 23, 2021 at 4:48 AM Tu, Lijuan <lijuan.tu at intel.com<mailto:lijuan.tu at intel.com>> wrote:


> -----Original Message-----
> From: Ali Alnubani <alialnu at nvidia.com<mailto:alialnu at nvidia.com>>
> Sent: 2021年3月23日 16:44
> To: Tu, Lijuan <lijuan.tu at intel.com<mailto:lijuan.tu at intel.com>>
> Cc: dts at dpdk.org<mailto:dts at dpdk.org>; dpdklab at iol.unh.edu<mailto:dpdklab at iol.unh.edu>
> Subject: RE: [dts] [PATCH] tests/nic_single_core_perf: modify
> accepted_tolerance to percentage
>
> > -----Original Message-----
> > From: dts <dts-bounces at dpdk.org<mailto:dts-bounces at dpdk.org>> On Behalf Of Lijuan Tu
> > Sent: Thursday, March 11, 2021 12:48 PM
> > To: dpdklab at iol.unh.edu<mailto:dpdklab at iol.unh.edu>
> > Cc: dts at dpdk.org<mailto:dts at dpdk.org>; Lijuan Tu <lijuan.tu at intel.com<mailto:lijuan.tu at intel.com>>
> > Subject: [dts] [PATCH] tests/nic_single_core_perf: modify
> > accepted_tolerance to percentage
> >
> > * accepted_tolerance in conf/nic_single_core_perf.cfg are changed to
> > percentage, e.g.
> >     accepted_tolerance = '1%%'
> >     if not set, it will be set to 2% by default.
> > * delat in json are chagned to percentage too, e.g.
> >     'delta': '1.5%'
> > * expected_throughput will be forced to 1Mpps for compute delat in
> > percentage
> > * add new conf test_interval in conf/nic_single_core_perf.cfg, which
> > indicates interval of getting throughput statistics in second, e.g.
> >     test_interval = 10
> >     if not set, it will be set to 5 seconds by default.
> > * collect better statistics , get multiple throughput statistics
> > during the test run, and compute the average as the final results.
> >
> > Signed-off-by: Lijuan Tu <lijuan.tu at intel.com<mailto:lijuan.tu at intel.com>>
> > ---
> <snipped>
> > +accepted_tolerance = '1%%'
> <snipped>
> > +                    diff =
> > + (self.throughput[fwd_config][frame_size][nb_desc] -
> > +
> > self.expected_throughput[fwd_config][frame_size][nb_desc])/ \
> > +
> > self.expected_throughput[fwd_config][frame_size][nb_desc]
>
> The value of `diff` here is still a decimal, and needs to be multiplied by a 100
> before it can be compared to `accepted_tolerance`.
>
> I suggest splitting this patch into multiple ones. One for making the percentage
> change and one for taking the mean for multiple readings since they are
> unrelated changes.
> I also see other unrelated changes not explained in the commit message.

Really appreciate for your review, I will accept your advices.

>
> Thanks,
> Ali
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mails.dpdk.org/archives/dts/attachments/20210326/9f89b817/attachment.htm>