[dpdk-dev] [PATCH v4 1/5] app/test-flow-perf: add flow performance skeleton

Andrew Rybchenko arybchenko at solarflare.com
Wed May 6 10:54:49 CEST 2020


On 5/6/20 11:51 AM, Wisam Monther wrote:
>
>> -----Original Message-----
>> From: Andrew Rybchenko <arybchenko at solarflare.com>
>> Sent: Wednesday, May 6, 2020 11:48 AM
>> To: Wisam Monther <wisamm at mellanox.com>; Ajit Khaparde
>> <ajit.khaparde at broadcom.com>
>> Cc: dpdk-dev <dev at dpdk.org>; Jack Min <jackmin at mellanox.com>; Thomas
>> Monjalon <thomas at monjalon.net>; Jerin Jacob <jerinjacobk at gmail.com>;
>> gerlitz.or at gmail.com; l.yan at epfl.ch
>> Subject: Re: [dpdk-dev] [PATCH v4 1/5] app/test-flow-perf: add flow
>> performance skeleton
>>
>> On 5/6/20 10:32 AM, Wisam Monther wrote:
>>> I agree, since this is a test application, We can have L4 UDP/TCP rss
>>> configuration, since all Flows/traffic are allowed here and there is
>>> nothing to be common here.
>> UDP RSS is less common and could simply be unsupported.
>> So, it will be harder to use the tool for corresponding NICs.
> Ok,
> So we can go with only IP without TCP
> This way we can call the support of hash up to L3 only.
> Are we ok with this?
>

I"m OK with IP+TCP as the most common case.

>>> From: Ajit Khaparde <ajit.khaparde at broadcom.com>
>>> Sent: Wednesday, May 6, 2020 5:50 AM
>>> To: Andrew Rybchenko <arybchenko at solarflare.com>
>>> Cc: Wisam Monther <wisamm at mellanox.com>; dpdk-dev
>> <dev at dpdk.org>; Jack
>>> Min <jackmin at mellanox.com>; Thomas Monjalon
>> <thomas at monjalon.net>;
>>> Jerin Jacob <jerinjacobk at gmail.com>; gerlitz.or at gmail.com;
>>> l.yan at epfl.ch
>>> Subject: Re: [dpdk-dev] [PATCH v4 1/5] app/test-flow-perf: add flow
>>> performance skeleton
>>>
>>>
>>>
>>> ::snip::
>>>
>>>
>>>> +             },
>>>> +             .rx_adv_conf = {
>>>> +                     .rss_conf.rss_hf =
>>>> +                                     ETH_RSS_IP  |
>>>> +                                     ETH_RSS_UDP |
>>> May be it is better to remove ETH_RSS_UDP by default, since it is less
>>> common that RSS for TCP because of possible fragmentation and packets
>>> from the same stream delivered to different CPU cores.
>>> If we want to enable RSS on L4 headers, then UDP and TCP should be fine.
>>> Its an example app anyway?
>>> Otherwise we can just stick with L3 hash like some of the other examples.
>>> ::snip::
>>>



More information about the dev mailing list