Using Bittorrent for Image Deployments to Bare Metal Nodes

When deploying images to bare metal nodes it is via a iSCSI mechanism. Essentially, a bare metal node presents a iSCSI target to the Openstack compute node that has the image and the compute node mounts the device and transfers the image unto the bare metal nodes disk, then the baremetal node is reset to use the image. This approach has proven itself valuable during the development stage of bare metal, but going forward where we must address scalability and performance thsi approach present problems. We have observed a couple of other approaches to this issue and of particular interest is large scale data center software update schemes where software or systems need to be update is a expeditious manner with care take as to not load the internal network. One such approach is Twitter’s use of Murder to perfbare metal nodes.orm their updating of their internal server, which are in the thousands. This blueprint addresses how a bittorrent can be implement to achieve the both scalability and performance when deploying images to a large number of bare metal nodes.

(lifeless) This will certainly want to be a blueprint when we come to do it. One complicating factor is that it will impinge on glance and ironic and possibly nova - it's definitely not a TripleO 'thing' although we'll probably end up funding it in terms of dev time. I'm going to obsolete this spec for now on that basis, as this is the wrong place to be doing the design work.