Baseline measurement of network latency impact on Zeebe 8.1

[lenaschoenburg]

Test with Zeebe 8.1.8 For a useful test, enable network message compression.

[lenaschoenburg]

First try: https://github.com/zeebe-io/zeebe-performance-test/actions/runs/4294706900

gh workflow run measure.yaml \
  -f name=os-baseline-81 \
  -f chaos=network-latency-5  \
  -f helm-arguments="--set camunda-platform.zeebe.image.tag=8.1.8 --set camunda-platform.zeebe-gateway.image.tag=8.1.8 --set zeebe.config.zeebe.broker.cluster.messageCompression=SNAPPY --set zeebe.config.zeebe.gateway.cluster.messageCompression=SNAPPY"

[lenaschoenburg]

Without network latency:

Metric	Run 1	Run 2	Run 3	Average
Throughput	127.762	127.490	124.160	126.471
Latency p99	3.597	4.818	4.539	4.318
Latency p90	1.777	2.080	2.348	2.068
Lantecy p50	0.739	0.833	1.015	0.862

With 5ms network latency:

Metric	Run 1	Run 2	Run 3	Average
Throughput	109.316	107.890	105.400	107.535
Latency p99	4.416	5.083	4.662	4.720
Latency p90	2.349	2.425	2.453	2.409
Lantecy p50	1.296	1.580	1.738	1.538

Conclusion

Increasing network latency from ~0ms to 5ms:

• Throughput drops by a factor of 0.85
• p99 Latency increases by a factor of 1.09
• p90 Latency increases by a factor of 1.16
• p50 Latency increases by a factor of 1.78