[Openstack-operators] Multi Host flag in Havana (neutron), central network node is perfect chocking point..
mali at awcloud.com
Fri Apr 11 02:46:02 UTC 2014
Yes, there are scheduler options in neutron.conf which can schedule network/router/load balancer to different agents. You also can check neutron/scheduler directory in Icehouse RC, you can find LeastRouterScheduler which will greatly help for resource usage. AFAIK, it is not backported to Havana, but easy to do it by hand.
By the way, if you distribute all the agents on all compute nodes, all your compute nodes are open to public, meaning that it will be assigned a public IP to make itself as a router gateway.
It also must be mentioning that when some agent fails, all the resources binded to that agent are down, and you have to re-schedule all of them by hand (command line interface). There's a script called neutron-ha-tool can help you deal with this situation. Check it in the following links:
----- Original Message -----
From: "Zeeshan Ali Shah" <zashah at kth.se>
To: openstack-operators at lists.openstack.org
Sent: 星期五, 2014年 4 月 11日 上午 4:08:15
Subject: [Openstack-operators] Multi Host flag in Havana (neutron), central network node is perfect chocking point..
There used to be a --multi_host flag in nova-network . i.e. to have in-out traffic from compute itself. I think that was more distributed and non-choking model .
We run a dedicated neutron node and public network goes in-out from it. With high traffic app this created a perfect single point of failure . And yes I have tested it with high traffic load.
I was thinking to run L3 and l2 agent on all compute nodes .
Q1. Does it makes sense ?
2. Has anyone done it ? We are using Havana release.
3 and any suggestion for above ? --
Zeeshan Ali Shah
System Administrator - PDC HPC
PhD researcher (IT security)
Kungliga Tekniska Hogskolan
+46 8 790 9115
OpenStack-operators mailing list
OpenStack-operators at lists.openstack.org
More information about the OpenStack-operators