Benchmark comparison

[mmolimar]

Is there any kind of comparison or benchmark between Confluent REST Proxy and Hermes?

[adamdubiel]

No, we did not compare ourselves to REST Proxy. However we just finished throughput measurements in bare Kafka 0.10 vs Hermes on our production cluster (we were switching clusters and had one we could experiment on), so we will probably publish those.

[mmolimar]

Ok, many thanks!

[adamdubiel]

Reopening till we actually do it :)

[prasanthmeenapalli]

Does anyone has performance metrics now

[anumadan]

Hi,

Has this benchmarking information been published anywhere? Please share the related information.

Thanks, Anu

[mareckmareck]

While we haven't done any public benchmarks nor comparisons to this day, we did do some internal testing of Hermes in various scenarios.

One of such scenarios that can shed some light on performance of Hermes is stress testing of Hermes frontends. During that test, a single instance of Hermes frontend of 8 VCPUs and 16GBs of RAM was used. It was then put under heavy load of traffic that increased with time. The traffic involved usage of around 1k topics.

This test showed that, in such configuration, a single instance of Hermes frontend is capable of easily dealing with traffic of ~6k RPS without much of a hassle, retaining stability and normal levels of functionality.

In addition to this test, we have tested Hermes infrastructure under a cumulatively bigger load (over 100k messages per second) and found out that Hermes can perform well and stable under such pressure.

Apart from that, it's worth mentioning that we have a module called hermes-benchmark that can be used to locally perform basic benchmarking of single instance publishing to Kafka, to see if it fits your needs. In this PR you can find example results for benchmark tests from laptop.

[druminski]

Keeping it open as it would be nice to create official documentation for hermes performance using information provided by @mareckmareck