Protecting NVMe/TCP PDU Data at 400 Gbps

The Storage Performance Development Kit (SPDK) provides a high-performance NVMe/TCP target that scales very well. As the SPDK NVMe/TCP target gains broader adoption, providing strong error detection using the data digest to protect the NVMe/TCP Protocol Data Units (NVMe/TCP PDUs) is very important. Can the SPDK community implement the data digest at high network throughputs efficiently? The community build the SPDK acceleration framework to accelerate data movement and transformation operation with hardware accelerators such as the Intel® Data Streaming Accelerator (Intel® DSA) for high-performance storage applications. On platforms without Intel® DSA, the acceleration framework uses the Intelligent Storage Acceleration Library (ISA-L) to reduce the computation cost of the data digest by vectorizing it. Join us to see the work that the SPDK community has done to analyze the performance and efficiency of the NVMe/TCP target with data digest enabled. We tested different workloads using ISA-L and Intel® DSA to compute the data digest at network throughputs of up to 400 Gbps, and demonstrated which workloads benefit from offloading the data digest computation to hardware accelerators.

• The Performance and Efficiency of NVMe/TCP at 400 gbps for block storage workloads.
• The computation cost of protecting NVMe/TCP Data PDU data at 400 Gbps with software-only implementations.
• Using ISA-L and hardware offloads to reduce the computation cost of data digest
• Which block storage workloads benefit the most by offloading the data digest computation to hardware.