The RDMA implementation currently batches posting of receive requests that are completed in a single poll, but it should wait and batch up to a specified number of RRs across multiple polls. This will reduce PCI traffic and help performance especially in multicore setups.
The RDMA implementation currently batches posting of receive requests that are completed in a single poll, but it should wait and batch up to a specified number of RRs across multiple polls. This will reduce PCI traffic and help performance especially in multicore setups.