Hi While multiple indexing jobs are running [all are map reduce jobs hard committing at the end of every mapper], the Old gen utilization spikes a lot and CMS aggressively tries to keep the mem usage < 70% of the heap [I set this to trigger CMS when heap is 70%]. CPU utilization is hardly around 30-40% during this time.
1) How does write affect JVM utilization in solr? I have an autocommit of 15 seconds with openSearcher = false. After a write is flushed to the disk, does the in mem version of the commit/docs still stick around in the heap? 2) What kind of settings can help improve write throughput? Should we increase the # of max Indexing threads (from 8 to match the # of cpu cores). 3) I have SSD and hence frequent commits are OK (compared to disks). Should we set aggressive auto commits if we want to thwart the memory pressure during writes? Any pointers would be great! Cheers Veera
