[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <4C87C710.8080804@gmail.com>
Date: Wed, 08 Sep 2010 19:25:36 +0200
From: Jarek Poplawski <jarkao2@...il.com>
To: Eric Dumazet <eric.dumazet@...il.com>
CC: Anand Raj Manickam <anandrm@...il.com>, netdev@...r.kernel.org,
netfilter-devel@...r.kernel.org, shemminger@...tta.com
Subject: Re: Kernel Panic on OOM with 10 HTB rules
Eric Dumazet wrote, On 09/08/2010 04:45 PM:
> Le mercredi 08 septembre 2010 à 19:39 +0530, Anand Raj Manickam a
> écrit :
>>
>> imq0 Link encap:UNSPEC HWaddr
>> 00-00-00-00-00-00-00-00-00-00-00-00-00-00-00-00
>> UP RUNNING NOARP MTU:16000 Metric:1
>> RX packets:129112 errors:0 dropped:0 overruns:0 frame:0
>> TX packets:129114 errors:0 dropped:0 overruns:0 carrier:0
>> collisions:0 txqueuelen:11000
>> RX bytes:31060964 (29.6 MiB) TX bytes:31062207 (29.6 MiB)
>>
> If you cannot switch to 64bit kernel, then you are forced to use lower
> queue lengths (I see your imq devices use insane 11000 txqueuelen)
>
> Each frame use 4K, maybe 16K, it depends on MTU.
>
> even if we dont take into account other needs :
> 11000 * 16K = 170 Mbytes per imqX
> 1000 * 4K = 4Mbytes per ethX
>
> 170M * 8 -> memory overflow
>
You should consider that htb creates by default one queue per class
with txqueuelen limit each. This should probably explain why your
problems start when you classify to many classes.
Jarek P.
--
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