mac address table is not tuned for big amount of nodes?

Bug #1528392 reported by Alexander Bozhenko
6
This bug affects 1 person
Affects Status Importance Assigned to Milestone
Mirantis OpenStack
New
Undecided
Unassigned

Bug Description

We have this article on our website:
https://www.mirantis.com/blog/improving-dhcp-performance-openstack/

It advice to set the following numbers:
net.ipv4.neigh.default.gc_thresh1 = 1024
net.ipv4.neigh.default.gc_thresh2 = 4096
net.ipv4.neigh.default.gc_thresh3 = 8192

I just checked 6.1 and 7.0 and those settings are not present there:

6.1:
sysctl -a | grep gc_thresh
net.ipv4.neigh.default.gc_thresh1 = 128
net.ipv4.neigh.default.gc_thresh2 = 512
net.ipv4.neigh.default.gc_thresh3 = 1024
net.ipv6.neigh.default.gc_thresh1 = 128
net.ipv6.neigh.default.gc_thresh2 = 512
net.ipv6.neigh.default.gc_thresh3 = 1024

In 7.0 looks like they are increased for ipv4 but gc_thresh2 and gc_thresh3 are 2 times lower than per advice in article:
sysctl -a | grep gc_thresh
net.ipv4.neigh.default.gc_thresh1 = 1024
net.ipv4.neigh.default.gc_thresh2 = 2048
net.ipv4.neigh.default.gc_thresh3 = 4096
net.ipv6.neigh.default.gc_thresh1 = 128
net.ipv6.neigh.default.gc_thresh2 = 512
net.ipv6.neigh.default.gc_thresh3 = 1024

Do we need to tune it for ipv4 in 6.1 in a MU?
Do we need to tune it for ipv6?

description: updated
Revision history for this message
Aleksandr Shaposhnikov (alashai8) wrote :

Hi Alex,
Thanks for reading the article :) Most of it outdated nowadays but mac table setting are still useful thing to tune if you're use neutron with "vanilla" agents like ovs. This settings are critical only if you have pretty big clusters with more than 128 VM's and became a really big issue when the number of VMs more than 1024. Speaking about IPv6 setting are also required though.
I didn't mentioned that in article because even right now IPv6 not so popular in deployments.

Revision history for this message
Alexander Bozhenko (alexbozhenko) wrote :

Hello, Aleksandr. Thank you for your prompt response.
Do you mean 128 vms on one compute or in one network, or in the whole cloud?

So it won't hurt to increase numbers for ipv6 in fuel as well, right? Ipv6 is used in production one one of the bug customers installation.

Revision history for this message
Denis Meltsaykin (dmeltsaykin) wrote :

Alexander B, we have already addressed the issue in https://bugs.launchpad.net/fuel/+bug/1488938 for 6.1. The fix will be available as a part of MU4. Marking this as Duplicate of 1488938.

To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.