[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20140123.144243.998328685775607324.davem@davemloft.net>
Date: Thu, 23 Jan 2014 14:42:43 -0800 (PST)
From: David Miller <davem@...emloft.net>
To: mkl@...gutronix.de
Cc: sojkam1@....cvut.cz, linux-can@...r.kernel.org,
netdev@...r.kernel.org
Subject: Re: [PATCH v2 1/2] can: Decrease default size of CAN_RAW socket
send queue
From: Marc Kleine-Budde <mkl@...gutronix.de>
Date: Thu, 23 Jan 2014 22:01:37 +0100
> On 01/23/2014 09:41 PM, David Miller wrote:
>> From: Michal Sojka <sojkam1@....cvut.cz>
>> Date: Wed, 22 Jan 2014 09:27:36 +0100
>>
>>> Since the length of the qdisc queue was set by default to 10
>>> packets, this is exactly what was happening.
>>
>> This is your bug, set the qdisc limit to something more reasonable.
>>
>> Something large enough to absorb the traffic wrt. the speed at which
>> the CAN device can sink the data.
>
> Hmmm, the problem is on i686 I have to increase the txqueuelen to 366
> before the socket works as expected (writes are blocking). This means if
> there are two processes one sending a stream of bulk data and the other
> one occasionally more important data there will be a lot of frames in
> front of the important ones.
>
> With Michal's patch we can limit the number of frames in flight to a
> reasonably low amount.
What do you think UDP applications have to do on slow interfaces?
If they care about overrunning the device queue and getting packet
drops, they themselves set a smaller socket send buffer size.
CAN is not special in this regard, please stop treating it as such.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists