[Chilli] Chilli Digest, Vol 63, Issue 14
David Bird
dbird at google.com
Sat Feb 28 00:28:01 UTC 2015
You can use --tcpmss to clamp the TCP MSS. But, I agree, your issue is very
likely gso related. Something like "sudo ethtool -K eth1 gro off"
On Fri, Feb 27, 2015 at 2:00 PM, Richard REY <richard at rexy.fr> wrote:
> Yes, but this doesn't explain why there isn't any problem when the flows
> pass through the appliance. this issue is seen only when the TCP flows is
> sent to an internal processes
>
> I'll try to disable packet off-loading next monday.
> I'll also try to forbid the "jumbo frame" on NIC
>
> Rexy
>
> Le 27/02/2015 20:50, Alexandru Gheorghe a écrit :
>
> Usually fragmentation needed means you need to adjust the MTU of your nic.
>>
>> Alex
>>
>>
>> On 02/26/2015 02:30 PM, Claus Stjernoe wrote:
>>
>>> Hi Richard,
>>>
>>> Try to disable packet off-loading on the Gigabit NIC.
>>>
>>> Kind regards,
>>> Claus
>>>
>>> -----Oprindelig meddelelse-----
>>> Fra: chilli-bounces at coova.org [mailto:chilli-bounces at coova.org] På
>>> vegne af chilli-request at coova.org
>>> Sendt: 26. februar 2015 13:00
>>> Til: chilli at coova.org
>>> Emne: Chilli Digest, Vol 63, Issue 14
>>>
>>> Send Chilli mailing list submissions to
>>> chilli at coova.org
>>>
>>> To subscribe or unsubscribe via the World Wide Web, visit
>>> http://lists.coova.org/cgi-bin/mailman/listinfo/chilli
>>> or, via email, send a message with subject or body 'help' to
>>> chilli-request at coova.org
>>>
>>> You can reach the person managing the list at
>>> chilli-owner at coova.org
>>>
>>> When replying, please edit your Subject line so it is more specific than
>>> "Re: Contents of Chilli digest..."
>>>
>>>
>>> Today's Topics:
>>>
>>> 1. Re: Chilli Digest, Vol 63, Issue 12 (Richard REY)
>>>
>>>
>>> ----------------------------------------------------------------------
>>>
>>> Message: 1
>>> Date: Wed, 25 Feb 2015 22:48:56 +0100
>>> From: Richard REY <richard at rexy.fr>
>>> To: chilli at coova.org
>>> Subject: Re: [Chilli] Chilli Digest, Vol 63, Issue 12
>>> Message-ID: <54EE4348.2090503 at rexy.fr>
>>> Content-Type: text/plain; charset=windows-1252; format=flowed
>>>
>>> Hi,
>>> I use COOVA in its last version (SVN revision 492) running on an ALCASAR
>>> appliance (mageia 4.1 - kernel 3.14.32), all is OK excepted a flow issue.
>>> When a TCP stream goes across the appliance there isn't any flow issue
>>> (forward mode). BUT when a TCP stream is sent inside the appliance (to a
>>> HTTP proxy for example) the observed flow is very weak (Input mode). The
>>> problem is maximal when the card connected to COOVA is in the Gigabit mode.
>>>
>>> Our investigation :
>>> From the appliance, we try to download a file located on a FTP server
>>> connected on the network controlled by COOVA. Without COOVA the observed
>>> flow in gigabit mode is about 888Mb/s whereas with COOVA we can see just
>>> 68 Kb/s (YES 68 Kb/s). Regarding the packets exchanged we can see ICMP
>>> packets sent by the appliance (Destination unreachable (Fragmentation
>>> needed).
>>> We found a by-pass ?solution? by fixing the speed of the card to
>>> 100Mb/s. At this speed the phenomenon is practically invisible (just one
>>> ICMP packet sent) and the flow approaches the maximum supported by the
>>> network card. On a virtual machine where we can't fix Ethernet speed, we
>>> must choose a 100Mb/s virtual Ethernet card when building the VM.
>>> We have the Pcap file of our investigation. I say it again, this issue
>>> is seen only when the TCP flows is sent to an internal processes (in
>>> 'forward' mode all is ok).
>>>
>>> Regards
>>> Rexy & ALCASAR team
>>>
>>> Le 23/02/2015 08:59, kl a ?crit :
>>>
>>>> I see problem on kernel version > 3.7 right up to 3.10. Above 3.10 not
>>>> tested.
>>>> However, in the other quoted post, it says kernel 2.6.39 upto 3.3.4.
>>>> Above that not tested.
>>>> Something not consist here.
>>>>
>>>> So which kernel version you guys are using which are seeing this
>>>> problem ?
>>>>
>>>>
>>>> On Monday, February 23, 2015 3:13 PM, Danny Magat <danny at sb-ps.ae>
>>>> wrote:
>>>> This problem is still exist on my setup.
>>>> http://lists.coova.org/pipermail/chilli/2013-November/002437.html
>>>>
>>>> I hope that somebody will give us a hint on how to solve this.
>>>>
>>>> Regards,
>>>> Danny
>>>>
>>>> -----Original Message-----
>>>> From: chilli-bounces at coova.org [mailto:chilli-bounces at coova.org] On
>>>> Behalf Of chilli-request at coova.org
>>>> Sent: Sunday, February 22, 2015 4:00 PM
>>>> To: chilli at coova.org
>>>> Subject: Chilli Digest, Vol 63, Issue 12
>>>>
>>>> Send Chilli mailing list submissions to
>>>> chilli at coova.org
>>>>
>>>> To subscribe or unsubscribe via the World Wide Web, visit
>>>> http://lists.coova.org/cgi-bin/mailman/listinfo/chilli
>>>> or, via email, send a message with subject or body 'help' to
>>>> chilli-request at coova.org
>>>>
>>>> You can reach the person managing the list at
>>>> chilli-owner at coova.org
>>>>
>>>> When replying, please edit your Subject line so it is more specific
>>>> than
>>>> "Re: Contents of Chilli digest..."
>>>>
>>>>
>>>> Today's Topics:
>>>>
>>>> 1. Too sharp upload bandwidth limit on 1gbps wired network (kl)
>>>>
>>>>
>>>> ----------------------------------------------------------------------
>>>>
>>>> Message: 1
>>>> Date: Sun, 22 Feb 2015 02:21:53 +0000 (UTC)
>>>> From: kl <kl522 at yahoo.com>
>>>> To: Chilli List <chilli at coova.org>
>>>> Subject: [Chilli] Too sharp upload bandwidth limit on 1gbps wired
>>>> network
>>>> Message-ID:
>>>> <122480453.6349415.1424571713923.JavaMail.yahoo at mail.yahoo.com>
>>>> Content-Type: text/plain; charset=UTF-8
>>>>
>>>>
>>>>
>>>> Nobody is facing this issue with coova chilli ?
>>>>
>>>> http://www.coova.org/node/5171
>>>>
>>>> https://github.com/coova/coova-chilli/issues/32
>>>>
>>>> Happens only on Gigabit wired network and certain kernel verson.
>>>> Applies to 1.2.9, 1.3.0 and other versions ....
>>>>
>>>>
>>>> ------------------------------
>>>>
>>>> _______________________________________________
>>>> Chilli mailing list
>>>> Chilli at coova.org
>>>> http://lists.coova.org/cgi-bin/mailman/listinfo/chilli
>>>>
>>>>
>>>> End of Chilli Digest, Vol 63, Issue 12
>>>> **************************************
>>>>
>>>>
>>>> _______________________________________________
>>>> Chilli mailing list
>>>> Chilli at coova.org
>>>> http://lists.coova.org/cgi-bin/mailman/listinfo/chilli
>>>> _______________________________________________
>>>> Chilli mailing list
>>>> Chilli at coova.org
>>>> http://lists.coova.org/cgi-bin/mailman/listinfo/chilli
>>>>
>>>>
>>> ------------------------------
>>>
>>> _______________________________________________
>>> Chilli mailing list
>>> Chilli at coova.org
>>> http://lists.coova.org/cgi-bin/mailman/listinfo/chilli
>>>
>>>
>>> End of Chilli Digest, Vol 63, Issue 14
>>> **************************************
>>> _______________________________________________
>>> Chilli mailing list
>>> Chilli at coova.org
>>> http://lists.coova.org/cgi-bin/mailman/listinfo/chilli
>>>
>>
>>
>>
>> _______________________________________________
>> Chilli mailing list
>> Chilli at coova.org
>> http://lists.coova.org/cgi-bin/mailman/listinfo/chilli
>>
>> _______________________________________________
> Chilli mailing list
> Chilli at coova.org
> http://lists.coova.org/cgi-bin/mailman/listinfo/chilli
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.coova.org/pipermail/chilli/attachments/20150227/0299adfe/attachment-0001.html>
More information about the Chilli
mailing list