Size-Aware Sharding for Improving Tail Latencies in In-Memory Key-Value Stores

Explore a conference talk that introduces size-aware sharding as a novel approach to improve tail latencies in in-memory key-value stores. Delve into the implementation of this concept in the Minos key-value store and understand how it addresses the critical issue of tail latencies in distributed applications with high fan-out ratios. Learn about the advantages of size-aware sharding over traditional size-unaware approaches like keyhash-based sharding, request dispatching, and stealing. Discover how Minos overcomes implementation challenges to maintain high throughput by utilizing hardware dispatch for small item requests and adaptive load balancing techniques. Compare Minos' performance to three state-of-the-art in-memory KV store designs, revealing significant improvements in 99th percentile latency and throughput under specific conditions.