Valkey 9.0 ElastiCache t3.small Client Latency

The Valkey 9.0 ElastiCache cache.t3.small benchmark on cache.t3.small now has a real load-generator-side latency signal. The earlier latency view was an empty placeholder, this run's ECS artifact includes Embedded Metric Format samples for ElastiCache/LoadGenerator / ClientLatency with p50, p99, and p99.9.

Run window

The report window is 2026-06-07 11:47:08 UTC to 2026-06-07 12:47:42 UTC. The latency series includes 3 TaskId series and 52 timestamp buckets with latency datapoints.

Latency summary

Metric Value Read
Client latency p50 0.691 ms Median load-generator-side latency sample.
Client latency p99 0.702 ms High-percentile aggregate latency sample.
Client latency p99.9 0.703 ms Highest aggregate percentile published for this series.
Worst stream p99 1.331 ms Highest per-stream p99 observed in the window.
Worst stream p99.9 1.375 ms Highest per-stream p99.9 observed in the window.
Benchmark average latency 0.654 ms memtier_benchmark report value.
Benchmark max latency 1.38 ms Maximum benchmark latency reported for the run.
Benchmark p95 latency 0.676 ms Benchmark-side p95 latency.
Benchmark p99 latency 0.677 ms Benchmark-side p99 latency.

Run reading

The normal aggregate p99 and p99.9 signal sits around 0.70 ms, which is low and stable for this run shape. The only notable early spike is the worst-stream range of roughly 1.33-1.38 ms at 2026-06-07 11:48:00 UTC.

By the end of the chart, the visible latency band sits around 0.72-0.86 ms. That keeps the run in a tight load-generator-side latency range instead of showing a missing or placeholder metric.

Measurement caveat

These are not direct per-request HDR histogram values. The EMF values come from latency measurements that memtier emits to stdout as rolling or average latency samples, then publishes as load-generator metrics. The p50, p99, and p99.9 values here are percentiles over those published EMF latency samples for the report window. That makes the result useful as an ECS/load-generator-side time series, while keeping it distinct from a request-level latency distribution.

Tags:
AWS, Amazon ElastiCache, valkey, Valkey 9.0, load testing, memtier_benchmark