Ceph is a leading choice of block and object storage for OpenStack-based Telco NFV deployments. Hyperconverged architectures are gaining popularity in this space, where Ceph client and OSD components compete for resources with VNF compute instances. NFV deployments aim at maximizing VNF packing, thus resource available for storage is precious, particularly Ceph RBD client, which we found to be a significant consumer of CPU resources on a Hyperconverged node. This motivated us to take an in-depth look at the Ceph RBD client datapath, which hasn't been a focus in the past. Join us to hear about this exploration, performance profiling howtos, our current findings and opportunities for performance optimizations as it relates to Telco NFV.
Target Audience is anyone who is looking to understand Ceph client better in the context of Telco NFV and hyperconverged infrastructure.