Hi Sundar, Have you considered writing your own custom filter for hosts as described in [1] ?
Thanks, Sourabh [1] http://docs.openstack.org/developer/nova/devref/filter_scheduler.html On Mon, Aug 31, 2015 at 9:58 AM, Sundar Nadathur <snada...@altera.com> wrote: > Hi all, > > I’d appreciate if the experts can point me in one direction or > another. If there are existing mechanisms, we don’t want to reinvent the > wheel. If there aren’t, I’d be interested in exploring clean ways to extend > and enhance nova scheduling. > > > > Thank you very much. > > > > Cheers, > > Sundar > > > > *From:* Sundar Nadathur > *Sent:* Monday, August 24, 2015 10:48 PM > *To:* 'openstack-dev@lists.openstack.org' > *Subject:* [Nova] Placing VMs based on multiple criteria > > > > Hi, > > Please advise me whether the following scenario requires changes to > nova scheduler or can be handled with existing scheduling mechanisms. > > > > I have a type of PCIe device (not necessarily a NIC or HBA). The device > can be configured with a set of user-defined resources – say A, B, C. Each > resource can be shared between a limited number of VMs -- say A can be > shared among 4 VMs, B among 8, etc. A VM image may request the need for a > specific list of features, say A and B. Then I want to place the VM on a > host according to these criteria: > > 1. If there are hosts with a PCIe device that already has A and B > configured, and has a free instance each of A and B, the VM must be placed > on one of those hosts. > > 2. Otherwise, find a host with this PCIe device that can be > configured with one instance each of A and B. > > > > It is not clear that this can be handled through 3rd party metadata. > Suppose we create host aggregates with properties like “resource=A” and > “resource=B”, and also associate properties like “resource=A” with VM > images. (A and B are UUIDs representing user-defined resources.) Perhaps Nova > scheduler can match the properties to select host aggregates that have all > properties that the VM requires. However: > > a. This would not be dynamic (i.e. track the free instances of each > resource), and > > b. This addresses only #1 above. > > > > Is there any way I can leverage existing scheduler mechanisms to solve > this VM placement problem? If not, do you have thoughts/comments on what > changes are needed? > > > > Thanks, and apologies in advance if I am not clear. Please feel free to > ask questions. > > > > Cheers, > > Sundar > > ------------------------------ > > Confidentiality Notice. > This message may contain information that is confidential or otherwise > protected from disclosure. If you are not the intended recipient, you are > hereby notified that any use, disclosure, dissemination, distribution, or > copying of this message, or any attachments, is strictly prohibited. If you > have received this message in error, please advise the sender by reply > e-mail, and delete the message and any attachments. Thank you. > > __________________________________________________________________________ > OpenStack Development Mailing List (not for usage questions) > Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev > >
__________________________________________________________________________ OpenStack Development Mailing List (not for usage questions) Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
