netdev - Re: [PATCH] net: can: Increase tx queue length

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] [day] [month] [year] [list]

Message-ID: <f29c891a-2e21-3b9f-83d8-7abd48031011@hartkopp.net>
Date:   Fri, 15 Mar 2019 14:53:04 +0100
From:   Oliver Hartkopp <socketcan@...tkopp.net>
To:     Appana Durga Kedareswara Rao <appanad@...inx.com>,
        Dave Taht <dave@...t.net>,
        Toke Høiland-Jørgensen <toke@...hat.com>,
        Andre Naujoks <nautsch2@...il.com>,
        "wg@...ndegger.com" <wg@...ndegger.com>,
        "mkl@...gutronix.de" <mkl@...gutronix.de>,
        "davem@...emloft.net" <davem@...emloft.net>,
        Martin Jerabek <ma.jer@...t.cz>
Cc:     "linux-can@...r.kernel.org" <linux-can@...r.kernel.org>,
        "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        Michal Sojka <sojkam1@....cvut.cz>
Subject: Re: [PATCH] net: can: Increase tx queue length

Hi Kedar,

On 3/15/19 11:04 AM, Appana Durga Kedareswara Rao wrote:

>> Having about 10 elements in a CAN driver tx queue allows to work with
>> queueing disciplines
>> (http://rtime.felk.cvut.cz/can/socketcan-qdisc-final.pdf) and also to maintain a
>> nearly real-time behaviour with outgoing traffic.
>>
>> When the CAN interface is not able to cope with the (intened) outgoing traffic
>> load, the applications should get an instant feedback about it.
>>
>> There is a difference between running CAN applications in the real world and
>> doing performance tests, where it makes sense to increase the tx-queue-len to
>> e.g. 1000 and dump 1000 frames into the driver to check the hardware
>> performance.
> 
> Thanks, Oliver, Martin, Andre, Toke, Dave for your inputs...
> So to conclude this the default txqueuelen 10 is ideal for real-time CAN traffic,
> For Stress/Performance tests user manually need to increase the txqueuelen based on his requirements.
> 
> Please correct me if my understanding is wrong.

Yes, I would confirm that approach.

As Martin Jerabek pointed to a discussion with Michal Sojka here:

https://marc.info/?l=linux-can&m=155222580602047&w=2

You might also go for a more academic view based on the number of 
different CAN applications on the host.

@Martin: Would you like to propose a patch for can.txt (now can.rst)?

Regards,
Oliver