So it turned out the DataStax java client round-robins servers by default, which made periodic huge latencies. Switching to DCAwareRoundRobinPolicy solved the problem.
Another question is how do you get the local DC name? The application can parse conf/cassandra-topology.properties manually, but since the server already knows which DC it belongs to, it would be nice that it can just specify local DC without giving actual names. Hobin On Sat, Oct 19, 2013 at 5:18 PM, Hobin Yoon <hobiny...@gmail.com> wrote: > I am experiencing huge latencies with a multi-data center Cassandra > cluster. With consistency level ONE, I expected almost the same latency > with the single data center setup. What could possibly affect the latency > in multi-data center setup? > > multi-DC setup > min max avg (ms) > 1 969 164.554264 > > single-DC setup > min max avg (ms) > 1 51 2.371786 > > I am using the Datastax Java client library ( > https://github.com/datastax/java-driver). > > This is the keyspace description. > > ~/work/cassandra/bin$ ./cqlsh `hostname` > Connected to cdbp at mdc-s70:9160. > [cqlsh 3.1.7 | Cassandra 1.2.9-SNAPSHOT | CQL spec 3.0.0 | Thrift protocol > 19.36.0] > Use HELP for help. > cqlsh> desc keyspace pbdp; > > CREATE KEYSPACE pbdp WITH replication = { > 'class': 'NetworkTopologyStrategy', > 'DC2': '1', > 'DC3': '1', > 'DC0': '1', > 'DC1': '1' > }; > > USE pbdp; > > CREATE TABLE tweet ( > tid bigint PRIMARY KEY, > created_at_rt bigint, > created_at_st text, > lati float, > longi float, > real_coord boolean, > sn text, > text_ text > ) WITH > bloom_filter_fp_chance=0.010000 AND > caching='KEYS_ONLY' AND > comment='' AND > dclocal_read_repair_chance=0.000000 AND > gc_grace_seconds=864000 AND > read_repair_chance=0.100000 AND > replicate_on_write='true' AND > populate_io_cache_on_flush='false' AND > compaction={'class': 'SizeTieredCompactionStrategy'} AND > compression={'sstable_compression': 'SnappyCompressor'}; > > Thanks, > Hobin >
