Is it possible to use OpenStack Neat on fully virtualized environments

141 views
Skip to first unread message

Albert Vonpupp

unread,
Sep 17, 2013, 5:30:31 PM9/17/13
to opensta...@googlegroups.com
Hello Anton,

Right now I only have virtualized resources to test Neat, it might not makes much sense right now to do that since Neat uses WOL.

So basically I have a single physical sever (shared with a couple more students) with 16Gb, so I could emulate there the physical machines using KVM. Each guest would be a nova-compute node of the OpenStack cloud.

I read on your thesis that the minimum size for an Ubuntu server image is 128Mb. To have those 28 machines I would need 3.5Gb for the 28 VMs and 512Mb for the virtualization overhead, so I guess that ~4Gb of RAM per physical (virtualized) guest should be enough, which lead me to have 2 (3 tops) KVM guests.

Did you run your experiments on physical machines? Do you think that the described method could work?

Is there a way for replacing WOL with anything else? I have seen this interesting project [1] but I still need to try it out.

[1] https://github.com/simoncadman/libvirt-wakeonlan

Many thanks.

Regards,
Albert.

Anton Beloglazov

unread,
Sep 24, 2013, 9:25:56 PM9/24/13
to opensta...@googlegroups.com
Hi Albert,

Sorry for taking long to reply. I've done my experiments on physical machines. I'm not entirely sure that live migration of emulated VMs can be done between VMs on a single physical server. If it's possible, the communication overhead would be much lower than migrating over network, so this setup won't completely represent a real infrastructure.

I suppose replacing WOL is possible, but it may require some modification of Neat's code.

Cheers,
Anton

Albert Vonpupp

unread,
Sep 24, 2013, 10:04:42 PM9/24/13
to opensta...@googlegroups.com
Thanks Anton,

No problem! Thanks for your help!

I switched to physical machines some days ago. I think I am almost done reproducing your experiments (I hope!), I just got the other error I described on a separated post.
Reply all
Reply to author
Forward
0 new messages