Hello,
I finished quick testing of two h-metrics instances running on the same host both using
the same external cassandra node.
Motivation for this was to find a bottleneck.
From previous testing we know following:
- with additional
h-metric instance the throughput was increased almost linearly
- question was why it's not possible to get higher throughput from single h-metrics
instance while we know that cassandra node is not the bottleneck
Test:
1 - run perf tests against 1 h-metrics with 4GB memory using one external cassandra node
2 - run perf tests against 2 h-metrics with 2GB memory each (on the same VM as previous
test) using one external cassandra node (again the same VM as previous test)
3 - compare results
Results:
- for small msgs (1 metrics per request) there was almost no difference -> CPU was
fully utilized -> for small msgs the CPU is bottleneck
- for bigger msgs (100 metrics per request) total throughput was almost 2 times better
-> it's possible to get 2 times better throughput from the same VM -> neither
cpu, nor memory, nor network is the bottleneck -> h-metrics is doing some kind of
throttling
- I tried 500 metrics per request as well but I was getting timout exceptions. Maybe it
was too much for this VM.
Filip