Joseph Suh presenting ideas in the Advanced Scheduling session of the OpenStack Essex Design Summit in Boston.

* conversation only visible if you're a member.

S. Crago, K. Dunn, P. Eads, L, Hochstein, D.-I. Kang, M. Kang, D. Modium, K., J. Suh, J. P. Walters, “Heterogeneous Cloud Computing,” Workshop on Parallel Programming on Accelerator Clusters, Austin, TX, September 2011.

Abstract: Current cloud computing infrastructure typically assumes a homogeneous collection of commodity hardware, with details about hardware variation intentionally hidden from users. In this paper, we present our approach for extending the traditional notions of cloud computing to provide a cloud-based access model to clusters that contain a heterogeneous architectures and accelerators. We describe our ongoing work extending the OpenStack cloud computing stack to support heterogeneous architectures and accelerators, and our experiences running OpenStack on our local heterogeneous cluster testbed. 

If you're interested in more details, Mikyung and Dong-In wrote up some slides with the proposed design, and there's a wiki page that aims to capture the different bare-metal scenarios that people might be thinking of. Oh, and come join us in the bare metal on nova Launchpad group.

The distributed scheduler functionality that's coming in Diablo looks like a nice mechanism for support heterogeneous architectures, by describing them using capabilties. Since the scheduler is still in active development, we can't use this as the basis of our implementation just yet. Luckily, all of the machinery for communicating capabilities from the compute node to the scheduler is already in the nova trunk, and there's nothing stopping us from using that machinery right now. 

Our plan is take our current architecture-aware scheduler, and modify it to use capability information that's periodically reported by the compute node. Once the distributed scheduler code finally drops, we should be able to transition our code to it relatively quickly.