linux-kernel - Re: [PATCH net] vmxnet3: use gro callback when UPT is enabled

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <06e4a534-d15d-4b17-b548-4927d42152e1@huawei.com>
Date:   Fri, 17 Mar 2023 10:37:22 +0800
From:   Yunsheng Lin <linyunsheng@...wei.com>
To:     Jakub Kicinski <kuba@...nel.org>, Ronak Doshi <doshir@...are.com>
CC:     "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        "stable@...r.kernel.org" <stable@...r.kernel.org>,
        Pv-drivers <Pv-drivers@...are.com>,
        "David S. Miller" <davem@...emloft.net>,
        Eric Dumazet <edumazet@...gle.com>,
        Paolo Abeni <pabeni@...hat.com>,
        Guolin Yang <gyang@...are.com>,
        open list <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH net] vmxnet3: use gro callback when UPT is enabled

On 2023/3/17 4:34, Jakub Kicinski wrote:
> On Thu, 16 Mar 2023 05:21:42 +0000 Ronak Doshi wrote:
>> Below are some sample test numbers collected by our perf team. 
>>                           Test                                    socket & msg size                          base               using only gro
>> 1VM    14vcpu UDP stream receive        256K 256 bytes (packets/sec)    217.01 Kps    187.98 Kps         -13.37%
>> 16VM  2vcpu   TCP stream send Thpt     8K     256 bytes (Gbps)                18.00 Gbps    17.02 Gbps         -5.44%
>> 1VM    14vcpu ResponseTimeMean Receive (in micro secs)                      163 us             170 us                -4.29%
> 
> A bit more than I suspected, thanks for the data.

Maybe we do some investigation to find out why the performace lost is more than
suspected first.

For example if LRO'ed skb is added in gro_list->list, and then new LRO'ed skb from
the same flow only go through the whole GSO processing only to find out we have to
flush out the old LRO'ed in the gro_list->list, and add new LRO'ed skb in gro_list->list
again?


> .
>