__consumer_offsets compaction settings / metadata performance

[Aminouvic]

Hi all, 

Currently with default settings 
min.cleanable.dirty.ratio/log.cleaner.min.cleanable.ratio (0.5) 
__consumer_offsets partition size grows up to 80Mb and this seems to slow down 
metadata related operations (notably commits / join group).
What are the normal/expected partition avg size for the __consumer_offsets 
topic ?
Are there any recommendations to settings 
min.cleanable.dirty.ratio/log.cleaner.min.cleanable.ratio for the 
__consumer_offsets topic ?

Kafka Cluster 
        • version 0.10.1
        • 6 nodes 
        • __consumer_offsets #partitions : 50 
        • Cleaner config :
                ○ min.cleanable.dirty.ratio : 0.5
                ○ log.cleaner.enable : True
                ○ log.cleaner.io.buffer.size : 524288
                ○ log.cleaner.dedupe.buffer.size : 1073741824
                ○ log.cleaner.io.max.bytes.per.second: 104857600

Regards,
Amine