Juju Powered Openstack + Ceph Networking Concerns

[James Beedy]

Hello,


First off let me compliment and thank you for creating such awesome open
source cloud/service orchestration software!! I am currently putting
together a Ceph backed Openstack deploy using Juju and am having difficulty
routing storage traffic between my compute node (running only nova-compute)
and my ceph-osd (running only ceph-osd) nodes. The issue I am having is
that storage traffic on my compute and storage nodes seems to be routed
through the 1G interfaces which are on the os-admin-network and my
ceph-public-network(os-admin-network == ceph-public-network). Having
storage traffic pass through the 1G interfaces and happen on the 1G
(os-admin-network/ceph-public-network) network is causing a bottleneck, and
is also concerning seeing as I wouldn’t expect to see the storage traffic
happen there. I am having extreme difficulty creating a separate network
and interfaces(using Juju) for compute to talk to ceph-osd on…..the only
resolve I can seem to arrive at is to manually configure a separate 10G net
and separate 10G interfaces for compute to talk to ceph-osd on (otherwise
my entire os-admin-network would have to be 10G) …which is difficult when
deploying with juju seeing as manual configuration is not really supported.


I have been baffled by this for a while now, not having my networks sorted
out is causing me to fall far behind schedule for getting out new stack up
and sending our servers to the datacenter. This of course is an imperative
configuration that I need to understand and/or implement correctly in order
to move forward with my deploy, if someone could share a little insight to
a configuration (juju configs) where ceph-osd and compute might talk on
~(os-admin-network) it would be greatly appreciated!


Thank you!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ubuntu.com/archives/juju/attachments/20150313/d0a965da/attachment.html>