I did not want the results to live only in CloudWatch. That works for running the test, but not so well for sharing or keeping the output around.

So I added the "Reporting" branch, and the first result is shown below.

That immediately exposed one issue: cache hit rate was not improving as expected. After checking the setup, it became clear why - the default item count, with that item size, was too large for cache.t4g.small. The test was not giving the cache enough room to reach the behavior I was expecting.

Avg Throughput

1,947ops/sec

Peak Throughput

7,422ops/sec

Throughput CV

41.3%

Avg Latency

1.32ms

Max Latency

2.40ms

Avg Bandwidth

287KB/s

Active Load Window

31min

Pre-fill Duration

29min

Avg ECS CPU

25.7%

Peak ECS CPU

55.3%

ECS Mem Peak

5MB

Loadgen Tasks

3

Avg Memory

67.16%

Max Memory

100.03%

Mem Headroom

-0.0%

Engine CPU Peak

14.6%

Cache Hit Rate

2.8%

Total Evictions

0

Min Free Mem

100MB

Peak Key Count

289,752

First Eviction

13:29 UTC

Memtier Benchmark

benchmark window only

Infrastructure

full CloudWatch window

ElastiCache Deep-Dive

full CloudWatch window