True. Ideally performance (and other) monitoring should be an optional service (module) with its own API etc. Wido's proposal does not preclude that design (for the future), it seems to me.
On 8/2/12 10:09 AM, "John Kinsella" <[email protected]> wrote: >Yeah this is fairly hugeŠI'm a KVM guy as well so can't comment on the >other platforms. I've been using an opsview/nagios module that tells me >when a disk has high IOPs, but we haven't added the code yet to look at >the individual VM yet. > >I guess the only concern that comes to mind is where's the line between >what your cloud management platform does and what your monitoring >platform does - don't really need repetition. > >John > >On Aug 2, 2012, at 8:23 AM, Wido den Hollander wrote: > >Hi, > >I've been thinking about this for a couple of months while working on RBD. > >In public clouds it's very important to know which user is (ab)using your >storage, so I figured it could be useful if CloudStack would also support >disk I/O polling. > >On the Wiki I started with a document about this: >http://wiki.cloudstack.org/pages/viewpage.action?pageId=11830451 > >I only have experience with KVM, so I'm not sure if and how other >Hypervisors support this. > >For KVM it seems the implementation is rather simple, but I rather not >make this a KVM-only feature. > >The first steps will be to lay the groundwork in CloudStack and the >polling for disk IOps and then implement it per Hypervisor. > >I'd like some input on other Hypervisors to get a better picture of what >has to be done. > >Comments and flames are welcome :) > >Wido > > >Stratosec<http://stratosec.co> - Secure Infrastructure as a Service >o: 415.315.9385 >@johnlkinsella<http://twitter.com/johnlkinsella> >
