In designing an OpenStack cloud system, one of the most critical tasks is how to build out a highly scalable and reliable storage system. We, at Adobe, have been using Ceph as high performance block storage in OpenStack for a variety of use cases such as images, volumes, and snapshots. Over the last two years, and through several experiences of building out Ceph systems, we've been able to achieve very competitive performance with traditional spinning HDDs, yet we found interesting bottlenecks on several subcomponents including networks. We would like to share how we designed our Ceph architecture to maximize its performance with all commodity hardware. We are also planning on enhancing our current Ceph architecture to further boost up its performance using NVMe SSDs.
In the second part of the talk, Intel will share with us the work they are doing with Ceph. Intel will cover their upstream development work, reference archiectures, and benchmarks including work on an all NVMe Ceph clusters. Then Intel will cover review initial results of their blustore testing as well as some early work with Ceph on 100GbE networks.