6.6.1 aggregation performance

Elastic Stack Elasticsearch
1

Hello,

I've built up 2 es testing clusters, one 6.3.2 and the other 6.6.1, both on identical hardware(16c64g+ssd).

6.6.1 runs on jdk11+g1gc
6.3.2 runs on jdk8 + cmsgc

After running "geonames" benchmark by esrally, I see a drop down
throughput in "country_agg_uncached" with 8% in 6.6.1 cluster, while other task results like term/phrase/script are nearly the same.

Cpu both reached 100% while running "country_agg_uncached" task in each cluster

Is there any implemetation change with aggregation happened? I've got confused by this performance drop down.

Any ideas?

2

Is there any possibility you can run those tests with the same JVM version and the same GC setup to remove those from the equation as being potential culprits? If you do, and find the same results, please open a github issue.

--Alex

3

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.