Whitepaper: RDMA for Ultra-Efficient, Low Latency 25Gb Ethernet
Eletter Product
High performance embedded systems can benefit from the use of RDMA (Remote Direct Memory Access) protocols to efficiently move large amounts of data with minimal latency and CPU loading. When combined with ultra-fast transport layers such as 25Gb Ethernet, RDMA can provide unmatched networking performance for embedded processing systems. This whitepaper addresses these topics:
- An overview of the RDMA/RoCE protocol
- An introduction to Critical I/O’s 25GbE XMC products
- A description of Critical I/O’s RDMA/RoCE driver support for Linux and VxWorks
- RDMA latency results: node-to-node message latencies as low as 3.5 usec
- RDMA rate results: up to 12GB/s (dual-port, bidirectional) @ 3% host CPU utilization
- A brief overview of the alternative UDP Direct protocol
Critical I/O’s networking products are designed to move the large amounts of data generated in high bandwidth systems, at full line rates, with ultra-low host CPU utilization and lowest possible latency.
