Originally from the User Slack
@FRED_FAT: Hi Gurus, recently I have migrated data from an old cluster to a new one, however I find the avg query (by primary key) performance is degrade around 30%, from 22ms to 26ms. Compare with other cluster(old cluster is destroyed) with exactly same data, there is no difference on hardware and network configuration. The only difference we see is the old cluster has more sstables (around 300 sstables each node) than other cluster (less than 100 sstables each node) . I am guessing whether query read more record version between multiple sstables which could cause more disk reads than normal cluster and I want to check further on it. But I have no idea how to check the avg disk reads for average or specific CQL from any internal virtual table or from tracing output, anyone has suggestion ? Or anyone know whether more sstables (without enough compaction) could cause additional disk reads ?
@avi: Try nodetool tablehistograms and tracing to see how many sstables are read per query
@FRED_FAT: thanks, I tested tracing with ‘bypass cache’ and can see some file io operation. will try nodetool tablehistograms