Thanks, Russell and Brian. Yes, that blueprint was initiated by USC-ISI team and I submitted the updated branch for the code review in June 2013. But, core reviewers don't want to see too limited approach for only GPU. So now we are going to work for LXC host device passthrough including GPU like KVM PCI passthrough, based on the review comments.
What we propose is to implement LXC host device passthrough. We want to be in sync with the current KVM pci-passthrough as much as possible. What we suggest is to utilize current nova/pci/* related stuffs and add LXC related codes there. If this is not good, a new separate patch for LXC host device needs to be done. I'll appreciate more opinions/comments. Anybody is welcome to collaboration. Thanks, Mikyung and David USC-ISI ----- Original Message ----- From: "Brian Schott" <[email protected]> To: "OpenStack Development Mailing List" <[email protected]> Sent: Thursday, August 22, 2013 9:21:46 AM Subject: Re: [openstack-dev] LXC host device passthrough Mikyung is on that USC-ISI team. I was going to ask the current maintainer JP if we should start retiring some of these older blueprints. GPU support has moved on since I started these in 2011 and maybe it makes sense to migrate to something that extends/derives from: https://blueprints.launchpad.net/nova/+spec/pci-passthrough-base https://blueprints.launchpad.net/openstack/?searchtext=pci-passthrough-libvirt I'd love to work on this again, but alas, small company, paying the bills.... Brian On Aug 22, 2013, at 8:30 AM, Russell Bryant < [email protected] > wrote: On 08/21/2013 05:31 PM, Mikyung Kang wrote: Hi, We are considering the LXC host device passthrough for GPU. Is anybody already working on this LXC host device passthrough? If it is not already done, we are interested in implementing this LXC host device passthrough. We want to get some opinions/comments about LXC host device passthrough. We think LXC host device passthrough is similar to KVM pci-passthrough, and can share much code for KVM PCI Passthrough: https://review.openstack.org/#/q/topic:pci-passthrough-enhancement,n,z We think it can be done by adding some code for LXC passthrough to the KVM pci-passthrough code by updating codes under nova/pci/ and creates a few more files. Maybe DB, scheduler, and object support wise, they can be adapted for LXC host device by adding one more table. Any better idea? Someone else has worked on this some. https://blueprints.launchpad.net/nova/+spec/heterogeneous-gpu-accelerator-support IIRC, the code still needed a good bit of work, though. -- Russell Bryant _______________________________________________ OpenStack-dev mailing list [email protected] http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev _______________________________________________ OpenStack-dev mailing list [email protected] http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev _______________________________________________ OpenStack-dev mailing list [email protected] http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
