There is work we're doing with HP regarding Quota management (see the blue print https://wiki.openstack.org/wiki/KeystoneCentralizedQuotaManagement). The aim behind this is to be able to define quotas at the domain level and then delegate the administration down to the domain managers within their area along with being able to allocate quotas to different cells/regions. It won't help for the spot market/utilisation scenarios though, more to provide a single pane of glass for quota management and delegation. Tim From: jonathan.proulx@gmail.com [mailto:jonathan.proulx@gmail.com] On Behalf Of Jonathan Proulx Sent: 26 July 2013 17:34 To: Tim Bell Cc: Di Pe; openstack-hpc@lists.openstack.org Subject: Re: [openstack-hpc] Looking for OpenStack workload management On Fri, Jul 26, 2013 at 5:33 AM, Tim Bell <Tim.Bell@cern.ch <mailto:Tim.Bell@cern.ch> > wrote: Where we are trying to find a good solution is regarding how to use spot market resources in our cloud. If we allocate out a quota to a project but they do not use the full resources, we want to be able to offer that quota to others on a low SLA (killed with little warning). This allows us to get our resource utilisation up while ensuring that projects can get the resources they're entitled to when there is a need. This is an interesting area to be getting a few sites together to share solutions. This is an area I'm looking at as well. My short term desire is to provide a way to provided for the same hardware sometimes being used to run virtualized instances and sometimes be provisioned as a baremetal system (either through Triple0 or manually reprovisioned). My short term solution seems to be evolving around host-aggregates and special flavors. That communicates the "these can disappear at any time" SLA but doesn't address the quota issue which I think is a more general need. In many deployment resource cost is expressed directly as cost, in my research environment we don't have any direct internal billing for compute resources and I don' t pretend to understand the labyrinth of grants that fund the place, so "cost" here is expressed in quotas. I want people to have large (possibly infinite) quotas for cheap resources like "spot instances" and small quotas for "expensive" resources for example aggregates with 1:1 virtual to physical resource allocation. I though I remembered talk of abstracting the quota system into it's own project at the Portland summit, but can't seem to find it in my notes or on line, if I didn't dream that I'd love it if someone could refresh my memory. -Jon Jonathan Proulx Sr. Technical Architect MIT CSAIL