Cassandra Anti-patterns

From http://www.slideshare.net/mattdennis/strangeloop-2012-apache-cassandra-anti-patterns

  1. Don't run Cassandra on a SAN.
    • SANs were designed to solve problems C* doesn't have.
  2. Don't have commit log and data on the same disk.
    • Commit log is sequential, data tends not to be.
  3. Careful with heap size.
    • Usually no more than 16GB.
  4. Low file handle limit.
    • Default tends to be 1024, but C* needs 32K-128K.
  5. You don't need a load-balancer.
    • It's a SPOF (Single Point Of Failure), plus C* doesn't have master nodes and clients will load-balance themselves.
  6. Don't read or write to a single node.
  7. Don't let your ring get unbalanced.
  8. Watch the partitioner, and don't use ByteOrdered.
  9. Watch the batch insert size, it must fit in memory!
  10. Use a 64-bit JVM.
    • Or else you cannot address BigData.
  11. Use the Sun JVM.
  12. Do not use Supercolumns.
  13. Use the Datastax Operations center.