Hello,

If you prefer to not play with MTU settings every time, I`d recommend to
recompile this kernel setting all GRE options as a module and build
openvswitch module from vendor package, it will work perfectly with 1.9+
at least.


On 01/07/2014 09:51 AM, Li, Chen wrote:
> eth4 MTU = 1500, and br-int MTU = 1200   =>  2.76 Gbits/sec
> 
> Also, I tested in Openstack environment, while all virtual ports MTU are 
> 1500, and eth4 MTU = 9000
> The bandwidth between  2 instances is about 2Gb/s.
> 
> Thanks.
> -chen
> 
> 
> -----Original Message-----
> From: Zhou, Han [mailto:hzh...@ebay.com] 
> Sent: Tuesday, January 07, 2014 1:46 PM
> To: Li, Chen; Chris Luke; 'Jesse Gross'
> Cc: discuss@openvswitch.org
> Subject: RE: [ovs-discuss] low bandwidth for gre
> 
> Hi Chen,
> 
> Are you sure it is only MTU change resulted in such a huge difference?
> It is hard to believe that fragmentation itself would lead to 9G -> 122K 
> performance degradation.
> Could you try eth4 MTU = 1500, and br-int MTU = 1200?
> 
> Best regards,
> Han
> 
> -----Original Message-----
> From: discuss-boun...@openvswitch.org 
> [mailto:discuss-boun...@openvswitch.org] On Behalf Of Li, Chen
> Sent: Tuesday, January 07, 2014 12:46 PM
> To: Chris Luke; 'Jesse Gross'
> Cc: discuss@openvswitch.org
> Subject: Re: [ovs-discuss] low bandwidth for gre
> 
> O.
> Thanks for the reply.
> I just tried... And it works great !! 
> I can achieve 9.84 Gbits/sec now while br-int has MTU 8000, while eth4 has 
> MTU 9000.
> 
> Thanks a lot for the help !
> 
> -chen
> 
> -----Original Message-----
> From: Chris Luke [mailto:chr...@flirble.org]
> Sent: Tuesday, January 07, 2014 11:46 AM
> To: Li, Chen; 'Jesse Gross'
> Cc: discuss@openvswitch.org
> Subject: RE: [ovs-discuss] low bandwidth for gre
> 
> No, he means the physical interface needs a larger MTU than the logical one. 
> The physical one has to transport the packets your logical interface accepts 
> plus the header data for GRE. If there isn't enough headroom then it must 
> split the packet in two and reassemble at the far end - a slow process.
> 
> Chris.
> 
> 
> 
> 
>> -----Original Message-----
>> From: discuss-boun...@openvswitch.org [mailto:discuss- 
>> boun...@openvswitch.org] On Behalf Of Li, Chen
>> Sent: Monday, January 06, 2014 22:23
>> To: Jesse Gross
>> Cc: discuss@openvswitch.org
>> Subject: Re: [ovs-discuss] low bandwidth for gre
>>
>> Sorry, I'm really new to network area.
>>
>> So, you mean I should keep the sender under the low MTU, while the 
>> receiver with higher MTU size ??
>>
>> But, how I can make sure the sender is always the sender ????
>>
>> Thanks.
>> -chen
>>
>> -----Original Message-----
>> From: Jesse Gross [mailto:je...@nicira.com]
>> Sent: Tuesday, January 07, 2014 11:20 AM
>> To: Li, Chen
>> Cc: discuss@openvswitch.org
>> Subject: Re: [ovs-discuss] low bandwidth for gre
>>
>> If you change the MTU of both interfaces then it will still result in 
>> fragmentation.
>>
>> On Mon, Jan 6, 2014 at 9:50 PM, Li, Chen <chen...@intel.com> wrote:
>>> Yes, fragmentation do affect the bandwidth.
>>>
>>> I run command  to change interfaces' MTU size to test if 
>>> fragmentation
>> makes a difference.
>>>         ifconfig eth4 mtu ${mtu_size}
>>>         ifconfig br-int mtu ${mtu_size}
>>>
>>> The default MTU = 1500    =>    122 Kbits/sec
>>>                     MTU = 3000    =>     419 Kbits/sec
>>>                     MTU = 6000    =>     780 Kbits/sec
>>>                     MTU = 9000    =>     1.14 Mbits/sec
>>>
>>>
>>> But, what I don’t understand is, I didn’t see really high CPU%.
>>>
>>> Also, there is no other way to improve the bandwidth ?
>>> Even under MTU=9000, 1.14 Mb/s vs. 10 Gb/s, the gap between the test
>> result we can get and the physical NIC is unacceptable!!
>>>
>>>
>>> Thanks.
>>> -chen
>>>
>>>
>>> -----Original Message-----
>>> From: Jesse Gross [mailto:je...@nicira.com]
>>> Sent: Tuesday, January 07, 2014 10:32 AM
>>> To: Li, Chen
>>> Cc: discuss@openvswitch.org
>>> Subject: Re: [ovs-discuss] low bandwidth for gre
>>>
>>> Well did you at least see if fragmentation makes a difference?
>>>
>>> On Mon, Jan 6, 2014 at 8:14 PM, Li, Chen <chen...@intel.com> wrote:
>>>> But 122 Kbits/s vs. 9.41Gb/s ??
>>>>
>>>> Is this correct ???
>>>>
>>>> How can I improve it ??
>>>>
>>>>
>>>> Thanks.
>>>> -chen
>>>>
>>>>
>>>> -----Original Message-----
>>>> From: Jesse Gross [mailto:je...@nicira.com]
>>>> Sent: Tuesday, January 07, 2014 12:24 AM
>>>> To: Li, Chen
>>>> Cc: discuss@openvswitch.org
>>>> Subject: Re: [ovs-discuss] low bandwidth for gre
>>>>
>>>> On Sun, Jan 5, 2014 at 11:58 PM, Li, Chen <chen...@intel.com> wrote:
>>>>> Hi list,
>>>>>
>>>>>
>>>>>
>>>>> I’m working under CentOS, so the kernel is 
>>>>> 2.6.32-358.123.2.openstack.el6.x86_64.
>>>>>
>>>>> openvswitch.x86_64      1.11.0_8ce28d-1.el6ost
>>>>>
>>>>>
>>>>>
>>>>> I have two physical machine in the test, each of them have a 10 Gb 
>>>>> NIC
>> card.
>>>>>
>>>>> The bandwidth between the two machine using iperf can achieve
>> 9.41Gb/s.
>>>>>
>>>>>
>>>>>
>>>>> After enable gre follow guide
>>>>> http://networkstatic.net/configuring-vxlan-and-gre-tunnels-on-open
>>>>> vs
>>>>> w
>>>>> i
>>>>> tch/#!prettyPhoto
>>>>>
>>>>> The bandwitdth between the two nodes are really low.
>>>>>
>>>>> Only  122 Kbits/s !!!!
>>>>
>>>> At a minimum, fragmentation is presumably occurring.
>> _______________________________________________
>> discuss mailing list
>> discuss@openvswitch.org
>> http://openvswitch.org/mailman/listinfo/discuss
> 
> _______________________________________________
> discuss mailing list
> discuss@openvswitch.org
> http://openvswitch.org/mailman/listinfo/discuss
> _______________________________________________
> discuss mailing list
> discuss@openvswitch.org
> http://openvswitch.org/mailman/listinfo/discuss
> 

_______________________________________________
discuss mailing list
discuss@openvswitch.org
http://openvswitch.org/mailman/listinfo/discuss

Reply via email to