Method and system for multi-path network operations between storage clusters

Abstract

Multi-Path approach to data migration between nodes in distributed data storage systems is rendered by creating multiple connections between application cores across interfaces on both an initiator node and a target node and load balancing whole messages and RDMA operations over the various pathways routed via the same or different network fabric types. The Multi-Path approach adapts to applications which are core-affine and those which are not. The balancing of whole messages/RDMA operations is performed by a send/receive API command specifying the target node for the whole message (and optionally a core number in the case of a core-affine target). The Multi-Path approach adapts dynamically to network errors by excluding faulty pathways, as needed, and shifting traffic to remaining healthy pathways with background recovery of pathways running periodically for self-healing. The Multi-Path also adapts dynamically to the load by diverting traffic from slow pathways and favoring pathways with normal load.