[Openstack] Cells: *how* experimental?

Clint Byrum clint at fewbar.com
Thu Jan 28 23:10:29 UTC 2016


Excerpts from Ken D'Ambrosio's message of 2016-01-28 12:39:01 -0800:
> Hey, all.  Trying to have geographically-dispersed legs to a ~60 node 
> liberty development cloud.  Based on the traffic I see back and forth on 
> our icehouse cloud, it makes me think that having a cell hierarchy for 
> the new cloud might cut down on the back-and-forth chatter -- which 
> would be good, as I don't want to be yelled at by the network team.  
> However, I see in the docs that it's been listed as "experimental" since 
> Juno, and three releases of "experimental" makes me wonder if it's 
> actually moving anywhere, and something I should even consider using, or 
> if I should use a different approach altogether (suggestions welcome).

OpenStack isn't designed to spread out a single "region" over high
latency, expensive WAN links.  You may want to split into two regions
if your facilities are far enough apart that they can't send quite a
lot of traffic between them.

However, if you have some requirement to have everything under that
one region, I can say that even in a 1000 hypervisor simulation I don't
see more than 100Mbit of traffic to the control plane that all of the
nodes share. I'd expect 30 nodes to be quite a bit less traffic. So at
that point the only extra traffic I'd expect to see would be north/south
traffic where the router is scheduled on one side of the link, and the
vms are on the other. You could also have a shared provider network
for external access, but you'd need to do some tricks to make that work
across an L3 link.




More information about the Openstack mailing list