Achieving Wire-Latency Storage Systems by Exploiting Hardware ACKs

Learn about Juneberry, a revolutionary low-latency communication framework for storage systems that achieves wire-latency performance by exploiting hardware ACKs in this 16-minute conference presentation from NSDI '25. Discover how this innovative approach differs from existing RPC frameworks by providing a fast path for storage requests that can be committed with a single round trip and server CPU bypass, delivering extremely low latency while executing requests asynchronously on the server CPU. Explore the Ordered Queue abstraction that leverages NICs' hardware ACKs as commit signals while ensuring system linearizability, and understand how Juneberry supports durability through persistent memory placement. Examine the implementation using commodity RDMA NICs and its integration into two real-world storage systems: Memcached (in-memory caching) and PMemKV (persistent memory-based key-value store). Review comprehensive evaluation results demonstrating significant latency improvements compared to traditional RPC approaches under write-intensive workloads, presented by researchers from Tsinghua University at the USENIX Networked Systems Design and Implementation conference.