Open Stack

On 01/22/2016 09:33 AM, Matthew Booth wrote:
> On Tue, Jan 19, 2016 at 8:47 PM, Fox, Kevin M <Kevin.Fox at pnnl.gov
> <mailto:Kevin.Fox at pnnl.gov>> wrote:
>
>     One feature I think we would like to see that could benefit from LVM
>     is some kind of multidisk support with better fault tolerance....
>
>     For example:
>     Say you have a node, and there are 20 vm's on it, and thats all the
>     disk io it could take. But say you have spare cpu/ram capacity other
>     then the diskio being used up. It would be nice to be able to add a
>     second disk, and be able to launch 20 more vm's, located on the
>     other disk.
>
>     If you combined them together into one file system (linear append or
>     raid0), you could loose all 40 vm's if something went wrong. That
>     may be more then you want to risk. If you could keep them as
>     separate file systems or logical volumes (maybe with contigous
>     lv's?) Each vm could only top out a spindle, but it would be much
>     more fault tolerant to failures on the machine. I can see some cases
>     where that tradeoff between individual vm performance and number of
>     vm's affected by a device failure can lean in that direction.
>
>     Thoughts?
>
>
> This is simple enough for a compute host. However, the real problem is
> in the scheduler. The scheduler needs to be aware that there are
> multiple distinct resource pools on the host. For e.g., if your 2 pools
> have 20G of disk each, that doesn't mean you can spin up an instance
> with 30G. This is the same problem the VMware driver has, and to the
> best of my knowledge there's still no solution to that.

The proposed solution is here:

https://review.openstack.org/#/c/253187/2/specs/mitaka/approved/generic-resource-pools.rst

Best,
-jay