Cephalocon 2023

Rook[1] spreads OSDs across the nodes of cluster to provide redundancy and availability. This is done using node topology labels[2], which are also added by Rook to the desired level in the CRUSH map[3]. Containerized workloads(pods)/clients using RBD storage are spread out across the nodes too. Currently, reads are served from the primary OSD of the PG which maybe located on a different node, zone or even region. This results in increased data transfer and latency, which has an impact on both cost and performance. Leveraging Ceph-CSI's read affinity feature[4], reads are served from the OSDs which are in proximity to the client. This will reduce data transfer and improve performance. This talk will cover the implementation, usage and benefits of read affinity for RBD volumes with Rook. [1] https://rook.io/ [2] https://github.com/rook/rook/blob/master/Documentation/CRDs/Cluster/ceph-cluster-crd.md#osd-topology [3] https://docs.ceph.com/en/latest/rados/operations/crush-map/#crush-map [4] https://github.com/ceph/ceph-csi/blob/devel/docs/deploy-rbd.md#read-affinity-using-crush-locations-for-rbd-volumes