[Openstack] How to enable jumbo frames for instances ?

Heiko Krämer info at honeybutcher.de
Mon Jan 27 08:53:45 UTC 2014


ahh ok,

you can try this one (https://bugs.launchpad.net/neutron/+bug/1075336)

and you can look here into the peace of code:
https://github.com/openstack/neutron/commit/5808a65fa12be096278f3793909a62bf1d79e82b


Maybe is that what you want.


Good luck
Heiko

On 27.01.2014 09:28, Li, Chen wrote:
> Hi Heiko,
>
> I do have tried edit dhcp-option-force in etc/neutron/dnsmasq.conf.
> Base on my test, the value dhcp-option-force=26,${MTU} is only affect
the MTU in instances, it not affected any virtual port that we found in
compute node.
> This means, even I set dhcp-option-force=26,9000, you can get MTU=9000
in instances, but due to the virtual ports in host is still have
MTU=1500, so the real package size instance will get is still 1500.
>
> Also, I do restricted instance's bandwidth by setting extra_specs in
flavor :
> nova flavor-show test_flavor
>
+----------------------------+-----------------------------------------------------------------------------------+
> | Property                   |
Value                                                                            
|
>
+----------------------------+-----------------------------------------------------------------------------------+
> | name                       | test_flavor        |
> | ram                        |
512                                                                              
|
> | OS-FLV-DISABLED:disabled   |
False                                                                            
|
> | vcpus                      |
1                                                                                
|
> | extra_specs                | {u'quota:vif_inbound_average':
u'51200', u'quota:vif_outbound_average': u'51200'} |
> | swap                      
|                                                                                  
|
> | os-flavor-access:is_public |
True                                                                             
|
> | rxtx_factor                |
1.0                                                                              
|
> | OS-FLV-EXT-DATA:ephemeral  |
0                                                                                
|
> | disk                       |
0                                                                                
|
> | id                         |
7                                                                                
|
>
+----------------------------+-----------------------------------------------------------------------------------+
>
> The reason I want to use jumbo frames is I hope jumbo frames will use
less CPU%, so that I can get higher bandwidth for a whole compute node.
>
> Thanks.
> -chen
>
>
> -----Original Message-----
> From: Heiko Krämer [mailto:info at honeybutcher.de]
> Sent: Monday, January 27, 2014 4:11 PM
> To: openstack at lists.openstack.org
> Subject: Re: [Openstack] How to enable jumbo frames for instances ?
>
> Hi Chen,
>
> first of all, the default supported network driver for kvm supported
only 1g.
>
> You can configure your MTU in your /etc/neutron/dnsmasq.conf on network
> node,like:
> dhcp-option-force=26,1454
>
> MTU = 1454
>
> You need to kill all dnsmasq processes on your network node and
restart dhcp agent.
>
> In addition you ought to consider that is maybe a bad idea to allow
each spawned instance to use the complete 10G connection. I'm using for
each compute node two 10G Nic's which is one for internal instance
communications and the other for storage attachments but every time only
1G to the instances.
>
> Maybe it helps.
>
>
> Cheers
> Heiko
>
> On 27.01.2014 08:39, Li, Chen wrote:
>> Hi list,
>>
>> I'm working under CentOS 6.4 + Havana + Neutron + OVS + gre.
>>
>> I'm testing performance for gre.
>>
>> I have a 10Gb/s NIC for compute Node.
>>
>> While, the max bandwidth I can get is small then 3Gb/s, even I have
> enough instances.
>> I noticed the reason the bandwidth can't reach higher is due to the
> utilization for one CPU core is already 100%.
>>
>> So, I want to try if I can get higher bandwidth if I have bigger MTU,
> because the default MTU = 1500.
>>
>> But, after I set network_device_mtu=8500 in "/etc/nova/nova.conf", and
> restart openstack-nova-compute service and re-create a new instance,
the MTU for devices is still 1500:
>>
>> 202: qbr053ac004-d6: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc
> noqueue state UNKNOWN
>>     link/ether da:c0:8d:c2:d5:1c brd ff:ff:ff:ff:ff:ff
>> 203: qvo053ac004-d6: <BROADCAST,MULTICAST,PROMISC,UP,LOWER_UP> mtu
> 1500 qdisc pfifo_fast state UP qlen 1000
>>     link/ether f6:0b:04:3f:9d:41 brd ff:ff:ff:ff:ff:ff
>> 204: qvb053ac004-d6: <BROADCAST,MULTICAST,PROMISC,UP,LOWER_UP> mtu
> 1500 qdisc pfifo_fast state UP qlen 1000
>>     link/ether da:c0:8d:c2:d5:1c brd ff:ff:ff:ff:ff:ff
>> 205: tap053ac004-d6: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc
> htb state UNKNOWN qlen 500
>>     link/ether fe:18:3e:c2:e9:84 brd ff:ff:ff:ff:ff:ff
>>
>> Anyone know why is this happen ?
>> How can I solve it ??
>>
>> Thanks.
>> -chen
>>
>>
>>
>> _______________________________________________
>> Mailing list:
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
>> Post to     : openstack at lists.openstack.org
>> Unsubscribe :
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
>
>
>
> _______________________________________________
> Mailing list:
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
> Post to     : openstack at lists.openstack.org
> Unsubscribe :
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
>






More information about the Openstack mailing list