[dpdk-dev] [dpdk-stable] [PATCH v4] mbuf: fix reset on mbuf free

Ferruh Yigit ferruh.yigit at intel.com
Thu Jan 21 10:15:22 CET 2021


On 1/19/2021 2:21 PM, Morten Brørup wrote:
>> From: Ferruh Yigit [mailto:ferruh.yigit at intel.com]
>> Sent: Tuesday, January 19, 2021 3:03 PM
>>
>> On 1/19/2021 12:27 PM, Morten Brørup wrote:
>>>> From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Ferruh Yigit
>>>> Sent: Tuesday, January 19, 2021 1:01 PM
>>>>
>>>> On 1/19/2021 8:53 AM, Morten Brørup wrote:
>>>>> Could someone at Intel please update the test script to provide
>>>> output according to the test plan? Or delegate to the right person.
>>>>>
>>>>> According to the test plan, the information requested by Olivier
>>>> should be in the test output already:
>>>>>
>>>>
>> http://git.dpdk.org/tools/dts/tree/test_plans/nic_single_core_perf_test
>>>> _plan.rst?h=next
>>>>>
>>>>> PS: I can't find out who is the maintainer of the test plan, so I'm
>>>> randomly pointing my finger at the test plan doc copyright holder.
>> :-)
>>>>>
>>>>
>>>> Hi Morten,
>>>>
>>>> Ali has a request to update the expected baseline, to be able to
>> detect
>>>> the
>>>> performance drops, let me internally figure out who can do this.
>>>>
>>>> And do you have any other request, or asking same thing?
>>>>
>>>
>>> Hi Ferruh,
>>>
>>> I am asking for something else:
>>>
>>> The test script does not provide the output that its documentation
>> says that it does.
>>>
>>> Apparently, the test script for nic_single_core_perf produces an
>> output table with these four columns (as seen at
>> https://lab.dpdk.org/results/dashboard/patchsets/15142/#env-18):
>>>
>>>      +--------+--------------------+-----------------------+----------
>> --------------------+
>>>      | Result | frame_size (bytes) | txd/rxd (descriptors) |
>> throughput Difference (Mpps) |
>>>      +--------+--------------------+-----------------------+----------
>> --------------------+
>>>      | PASS   | 64                 | 512                   | 1.57100
>> |
>>>      +--------+--------------------+-----------------------+----------
>> --------------------+
>>>      | PASS   | 64                 | 2048                  | 1.87500
>> |
>>>      +--------+--------------------+-----------------------+----------
>> --------------------+
>>>
>>> But the test plan documentation (at
>> http://git.dpdk.org/tools/dts/tree/test_plans/nic_single_core_perf_test
>> _plan.rst) says that this output should be produced:
>>>
>>>      +------------+---------+-------------+---------+-----------------
>> ----+
>>>      | Frame Size | TXD/RXD |  Throughput |   Rate  | Expected
>> Throughput |
>>>      +------------+---------+-------------+---------+-----------------
>> ----+
>>>      |     64     |   512   | xxxxxx Mpps |   xxx % |     xxx    Mpps
>> |
>>>      +------------+---------+-------------+---------+-----------------
>> ----+
>>>      |     64     |   2048  | xxxxxx Mpps |   xxx % |     xxx    Mpps
>> |
>>>      +------------+---------+-------------+---------+-----------------
>> ----+
>>>
>>> Olivier and I am saying that only showing the Throughput Difference
>> (Mpps) does not provide any perspective to the result.
>>>
>>> I am requesting that the Expected Throughput (Mpps) should be shown
>> in the result too, as documented in the test plan.
>>>
>>
>> Ahh, this has a history, when the initial community lab infrastructure
>> prepared
>> some vendor(s) didn't want to show the actual throughput numbers.
>>
>> That is why this diff and baseline introduced, and this is the how
>> current
>> infrastructure works. So this is not something related to Intel.
>>
>> And as you can imagine this is not a technical issue, some companies
>> seems not
>> willing to share their performance numbers via community lab, and I
>> don't know
>> if something changed here in last a few years.
>>
> 
> That explains it!
> 
> If those companies still want to keep the community lab performance data hidden (which I don't object to), wouldn't it be better if the performance test scripts output the deviation from the expected throughput in percent (with one or two decimals after the comma) instead of in Mpps?
> 

Sounds reasonable, I assume there is a reason behind it but I don't remember, 
cc'ed lab.


>>
>>>>>
>>>>> Med venlig hilsen / kind regards
>>>>> - Morten Brørup
>>>>>
>>>>>> -----Original Message-----
>>>>>> From: Olivier Matz [mailto:olivier.matz at 6wind.com]
>>>>>> Sent: Tuesday, January 19, 2021 9:32 AM
>>>>>> To: Ali Alnubani
>>>>>> Cc: David Marchand; Ferruh Yigit; zhaoyan.chen at intel.com; dev;
>>>> Andrew
>>>>>> Rybchenko; Ananyev, Konstantin; Morten Brørup;
>>>> ajitkhaparde at gmail.com;
>>>>>> dpdk stable; Ajit Khaparde; Slava Ovsiienko; Alexander Kozyrev
>>>>>> Subject: Re: [dpdk-stable] [PATCH v4] mbuf: fix reset on mbuf free
>>>>>>
>>>>>> Hi Ali,
>>>>>>
>>>>>>
>>>>>> On Mon, Jan 18, 2021 at 05:52:32PM +0000, Ali Alnubani wrote:
>>>>>>> Hi,
>>>>>>> (Sorry had to resend this to some recipients due to mail server
>>>>>> problems).
>>>>>>>
>>>>>>> Just confirming that I can still reproduce the regression with
>>>> single
>>>>>> core and 64B frames on other servers.
>>>>>>
>>>>>> Many thanks for the feedback. Can you please detail what is the
>>>> amount
>>>>>> of performance loss in percent, and confirm the test case? (I
>>>> suppose
>>>>>> it
>>>>>> is testpmd io forward).
>>>>>>
>>>>>> Unfortunatly, I won't be able to spend a lot of time on this soon
>>>>>> (sorry
>>>>>> for that). So I see at least these 2 options:
>>>>>>
>>>>>> - postpone the patch again, until I can find more time to analyze
>>>>>>      and optimize
>>>>>> - apply the patch if the performance loss is acceptable compared
>> to
>>>>>>      the added value of fixing a bug
>>>>>>
>>>>>> Regards,
>>>>>> Olivier
>>>>>>
>>>>>>
>>>>>>>
>>>>>>> - Ali
>>>>>>>
>>>>>>>> -----Original Message-----
>>>>>>>> From: Ali Alnubani <alialnu at nvidia.com>
>>>>>>>> Sent: Friday, January 15, 2021 8:39 PM
>>>>>>>> To: David Marchand <david.marchand at redhat.com>; Olivier Matz
>>>>>>>> <olivier.matz at 6wind.com>; Ferruh Yigit <ferruh.yigit at intel.com>;
>>>>>>>> zhaoyan.chen at intel.com
>>>>>>>> Cc: dev <dev at dpdk.org>; Andrew Rybchenko
>>>>>>>> <andrew.rybchenko at oktetlabs.ru>; Ananyev, Konstantin
>>>>>>>> <konstantin.ananyev at intel.com>; Morten Brørup
>>>>>>>> <mb at smartsharesystems.com>; ajitkhaparde at gmail.com; dpdk stable
>>>>>>>> <stable at dpdk.org>; Ajit Khaparde <ajit.khaparde at broadcom.com>
>>>>>>>> Subject: Re: [dpdk-stable] [PATCH v4] mbuf: fix reset on mbuf
>> free
>>>>>>>>
>>>>>>>> Hi,
>>>>>>>> Adding Ferruh and Zhaoyan,
>>>>>>>>
>>>>>>>>> Ali,
>>>>>>>>>
>>>>>>>>> You reported some performance regression, did you confirm it?
>>>>>>>>> If I get no reply by monday, I'll proceed with this patch.
>>>>>>>>
>>>>>>>> Sure I'll confirm by Monday.
>>>>>>>>
>>>>>>>> Doesn't the regression also reproduce on the Lab's Intel
>> servers?
>>>>>>>> Even though the check iol-intel-Performance isn't failing, I can
>>>>>> see that the
>>>>>>>> throughput differences from expected for this patch are less
>> than
>>>>>> those of
>>>>>>>> another patch that was tested only 20 minutes earlier. Both
>>>> patches
>>>>>> were
>>>>>>>> applied to the same tree:
>>>>>>>>
>>>>>>>> https://mails.dpdk.org/archives/test-report/2021-
>>>>>> January/173927.html
>>>>>>>>> | 64         | 512     | 1.571                               |
>>>>>>>>
>>>>>>>> https://mails.dpdk.org/archives/test-report/2021-
>>>>>> January/173919.html
>>>>>>>>> | 64         | 512     | 2.698                               |
>>>>>>>>
>>>>>>>> Assuming that pw86457 doesn't have an effect on this test, it
>>>> looks
>>>>>> to me
>>>>>>>> that this patch caused a regression in Intel hardware as well.
>>>>>>>>
>>>>>>>> Can someone update the baseline's expected values for the Intel
>>>>>> NICs and
>>>>>>>> rerun the test on this patch?
>>>>>>>>
>>>>>>>> Thanks,
>>>>>>>> Ali
>>>>>
>>>>
>>>
>>
> 



More information about the dev mailing list