Notes from production.

How long does a Redis Sentinel failover actually take? I measured it

I killed Redis, Valkey, and Dragonfly primaries 15 times and measured how long writes actually fail before Sentinel promotes a replica. Same answer for all three engines — runnable Docker lab included.

16 Sep 2026

I benchmarked KeyDB and Garnet too — one hit a wall, one broke my assumptions

I ran my benchmark harness against the two in-memory stores almost nobody tests independently — KeyDB and Microsoft's Garnet. KeyDB hit a hard wall at 16 threads; Garnet scaled past 21M ops/s, beating Dragonfly's number on half the cores.

13 Sep 2026

Redis vs Valkey vs Dragonfly vs KeyDB vs Garnet: every number, on 48 bare-metal cores

I benchmarked Dragonfly, Redis, and Valkey on 48 real bare-metal cores. Not "X wins" — a quantified trade-off between raw throughput and operational simplicity, with the reproducible harness and raw data.

04 Sep 2026

I benchmarked Dragonfly vs Redis vs Valkey. First, let me show you how I kept it honest.

Before trusting the benchmark: the traps that make most in-memory benchmarks lie — measuring the client, the VM, or a cache-miss path — and exactly how I disarmed each, including a bias I caught in my own setup.

04 Sep 2026

Dragonfly vs Redis vs Valkey: throughput and scaling on 48 real cores

Dragonfly vs Redis vs Valkey on 48 real cores — full setup, every tweak, and the scaling chart where three different shapes appear. Dragonfly ends 3.4× ahead of a realistic cluster.

04 Sep 2026

Why a normal client gets 4.6M ops/s out of a Redis cluster that can do 40M

A debugging story that became the study's biggest finding: why a normal client gets 4.6M ops/s from a Redis cluster that can actually do 40M — and what it means for running Redis Cluster in production.

04 Sep 2026

Latency under load: the same story as throughput, from the other side

Latency at 48 cores: Dragonfly stays sub-millisecond while a naive cluster client balloons to 84ms p50 — and why that's client-side queuing, not the shards, read honestly.

04 Sep 2026

Memory efficiency: bytes per key

The least-disputable number in the series: Dragonfly stores a million keys in ~13% less memory than Redis or Valkey — identical on a laptop and on bare metal.

04 Sep 2026

Operational simplicity: the cross-key tax, quantified

The cross-key tax, quantified: Redis Cluster rejects multi-key ops (MGET, transactions, Lua) unless keys share a slot; Dragonfly does them across the whole keyspace on one node.

04 Sep 2026

When to use which: Dragonfly vs Redis vs Valkey

No universal winner: a decision table across throughput, operational simplicity, failure domain, memory, and licensing — and the honest one-liner on when to pick Dragonfly vs a cluster.

04 Sep 2026