01 home
// featured · system design
Your Cache Proxy Layer Runs 4x Slower and Costs 3x More
A QCon benchmark of Envoy-proxied Valkey versus cluster-aware direct access: 2.4ms versus 567 microseconds at the tail, one third the cost for the same throughput, and one slow shard that takes down the whole cluster when a proxy layer shares the connection pools.
$
2026.10.30 · 5 min read · #system design
