Remote direct memory access (RDMA) is a well-known technology at the heart of the world’s fastest supercomputers and largest data centers. In short, RDMA is a remote memory-management capability that enables server-to-server data movement directly between application memories without CPU involvement. Offloading data movement from the CPU will result in performance and efficiency gains, while also significantly reducing latency. RDMA first became widely adopted in the High Performance Computing (HPC) industry with InfiniBand, but is now being leveraged by cloud, storage, and enterprise Ethernet networks with RDMA over Converged Ethernet (RoCE).
Given its broad expertise in RDMA technology, the InfiniBand Trade Association (IBTA) developed the RoCE standard and released the first specification in 2010. Although RoCE has been well-received by the enterprise storage and networking industry, especially by those wanting to accelerate application performance without overhauling their existing Ethernet infrastructure, there’s still some misinformation that continues on about the technology. Read on as we lay out and address the 11 most common myths surrounding RoCE.
1. RoCE requires a lossless network.
Initial deployments of RoCE required configuring the network to be lossless. However, the most advanced implementations of RoCE are resilient to packet loss and are able to run over ordinary Ethernet networks without the need for priority-based flow control. This resilient RoCE enables cloud, storage, and enterprise customers to deploy RoCE more quickly and easily while accelerating application performance, improving total infrastructure efficiency, and reducing cost.
2. RoCE doesn’t scale.
RoCE is currently deployed within Microsoft Azure Cloud, one of the largest cloud service providers, connecting tens of thousands of their compute and storage nodes.