4 ms·

The main selling point is that Cassandra has replaced HDFS as the underlying filesystem for Hadoop. I'm curious: what are the performance implications of using

by ajays 15y ago

The main selling point is that Cassandra has replaced HDFS as the underlying filesystem for Hadoop. I'm curious: what are the performance implications of using CFS instead of HDFS? Is it faster/slower/more fault-tolerant? (It appears to be the latter, given that there's no single point of failure in CFS).

jbellis 15y ago
That's one point: CFS means it's more fault tolerant and simpler to run (only one type of node), with equal-or-better performance. You also get multi-datacenter support essentially for free. The other big win is you can run analytic queries against your realtime data (data in Cassandra columnfamilies/tables, rather than a blob interface like CFS/HDFS) instantly, with no ETL. (An example: http://www.datastax.com/docs/0.8/brisk/brisk_demo. http://www.datastax.com/docs/0.8/brisk/brisk_demo.) Finally, Brisk takes care of integrating the Hadoop job/task trackers with Cassandra automagically, which otherwise requires fairly deep knowledge of both Hadoop and Cassandra (http://wiki.apache.org/cassandra/HadoopSupport http://wiki.apache.org/cassandra/HadoopSupport). Brisk also integrates the full stack (including Hive queries) with DataStax OpsCenter: https://s3.amazonaws.com/uploads.hipchat.com/6528/23268/uohlm21erckytzx/jt.png https://s3.amazonaws.com/uploads.hipchat.com/6528/23268/uohl...