6

With traditional RDBMS we are used to RAID10 in most cases but if using cassandra RF=2 then we exactly have one copy as backup then in this case why not or why to use RAID10.

I think it will reduce overhead from cassandra for replication..

Moreover in RAID10 if a hard drive fail then whole node will keep on working but if replication is used then one hard drive failure would result in whole node will down?

Though I think using RAID10 there will be overhead on each write but flushing is done when SSTABLE is full so it will not be felt all the time..

Gary Lindahl
  • 5,341
  • 2
  • 19
  • 18

1 Answers1

6

I would argue that RAID 10 is a waste of money. Two reasons:

1) One of the important attributes of BigTable (Cassandra or HBase) is the ability to quickly and cheaply expand your cluster or add redundancy by adding new servers. Based on recent prices, RAID 10 (striping AND spanning) is so expensive that it is virtually the same price as adding another whole server with JBOD storage.

2) Cassandra replication protects you from machine failure, not just disk failure. RAID 10 won't protect you if your CPU dies, but Cassandra replication will do. It will also protect you from disk failure, and will allow multiple clients to read from multiple nodes, preventing hotspots.

Chris Shain
  • 50,833
  • 6
  • 93
  • 125