Hi Alain, Thank you for your response. In my case, counter is the main table, having almost 40% of all data.
Thank you for the recommendation about testing on one node. 2018-01-18 13:02 GMT+02:00 Alain RODRIGUEZ <arodr...@gmail.com>: > Hello, > > I believe there is not a really specifically good strategy for counters. > Often counter tables size is relatively low (compared to events / raw data). > So depending on the workload you might want to pick one or the other. Given > the high number of reads the table will have to face (during reads + > writes), LCS might be a good choice if there is no better reason to pick > another strategy. Be aware that LCS have the highest write amplification > (Data will be written about 8 times on disk through compaction process) but > should be in a very nice spot for reads, mostly touching one or a few > SSTables. > > In the past I did not care much about the compaction strategy for counters > as I considered it to be negligible in my case (counters were MB big tables > out of a few TB for the entire dataset). > > You can always pick a strategy you think would work better, and test the > change on a canary node (use JMX to apply on 1 node only), see how it goes. > I found the doc for this on Datastax website. I hope this will help: > https://support.datastax.com/hc/en-us/articles/213370546-Change-CompactionStrategy-and-sub-properties-via-JMX > > C*heers, > ----------------------- > Alain Rodriguez - @arodream - al...@thelastpickle.com > France / Spain > > The Last Pickle - Apache Cassandra Consulting > http://www.thelastpickle.com > > 2018-01-17 16:14 GMT+00:00 Octavian Rinciog <octavian.rinc...@gmail.com>: >> >> Hello! >> I am using Cassandra 3.10. >> I have a counter table, with the following schema and RF=1 >> >> CREATE TABLE edges ( >> src_id text, >> src_type text, >> source text >> weight counter, >> PRIMARY KEY ((src_id, src_type), source) >> ); >> >> SELECT vs UPDATE requests ratio for this table is 0.1 >> READ vs WRITE rate, given by iostat is 100:1. >> Counter cache hit rate is 80%, so only for 20% UPDATE requests, the >> hard-disk is touched. >> >> I want to ask you which compation strategy is best for this table >> (SizeTieredCompactionStrategy or >> LeveledCompactionStrategy). >> >> Thank you, >> -- >> Octavian Rinciog >> >> --------------------------------------------------------------------- >> To unsubscribe, e-mail: user-unsubscr...@cassandra.apache.org >> For additional commands, e-mail: user-h...@cassandra.apache.org >> > -- Octavian Rinciog --------------------------------------------------------------------- To unsubscribe, e-mail: user-unsubscr...@cassandra.apache.org For additional commands, e-mail: user-h...@cassandra.apache.org
