Hi Yifan,

Try information from this thread.

Thanks,
Daniel Benli Ye

> Begin forwarded message:
> 
> From: benli ye <[email protected]>
> Subject: Re: net/mlx5 statistics related
> Date: August 5, 2019 at 10:30:41 AM GMT+8
> To: Yongseok Koh <[email protected]>
> Cc: "[email protected]" <[email protected]>, Ori Kam <[email protected]>, Dekel 
> Peled <[email protected]>, Slava Ovsiienko <[email protected]>, 
> Shahaf Shuler <[email protected]>
> 
> Hi Yongseok,
> 
> Thanks for replying. I will have a try.
> 
> Thanks,
> Daniel Benli Ye
> 
> 
>> On Aug 3, 2019, at 5:53 AM, Yongseok Koh <[email protected]> wrote:
>> 
>> Hi,
>> 
>> Can you try the following patch, which had been added after 18.11.0
>> I'll try to push it for 18.11.3.
>> 
>> http://git.dpdk.org/dpdk/commit/?id=ce9494d76c478345679abbda2a5606646de9886f
>> 
>> 
>> Thanks,
>> Yongseok
>> 
>>> On Jul 28, 2019, at 2:12 AM, benli ye <[email protected]> wrote:
>>> 
>>> Dear Mellanox experts,
>>> 
>>> I am running DPDK 18.11.0 performance test with mlx5 PMD driver on device 
>>> ConnectX®-4 Lx EN. Our own DPDK app (a Load Balance app) is used and UDP 64 
>>> bytes is generated by tester.
>>> We found that when we generated about 2.6Mpps UDP traffic for 1 core and 2 
>>> port setting testbed, our app starts to drops packets. However, the strange 
>>> issue is we cannot see any imissed or error statistics from DPDK layer.
>>> I saw the Mellanox test performance result in 
>>> https://fast.dpdk.org/doc/perf/DPDK_18_11_Mellanox_NIC_performance_report.pdf
>>>  and how do you find packet lose? Is there a bug in statistic calculation 
>>> in DPDK or there is another way to trace packet drop?
>>> Retrieve DPDK statistics is also used for Intel NIC and the behavior is 
>>> right. it should not be wrong for our usage in our own app.
>>> 
>>> Any suggestions would be welcomed. Thanks very much!
>>> 
>>> 
>>> Bests,
>>> Daniel Benli Ye
>> 
> 

Reply via email to