Hosting an HBase cluster for few weeks now, I'm facing some performances issues when it's time to run big MapReduce jobs. I have started to add more CPUs, more memory, etc. But based on a discussion on the HBase list it came out that disks performances might be an issue to. Few options came into the discussion. RAID, JBOD and LVM. Since I have access to a RAID controller and have 2 brand new 2TB hard drives I have decided to run some performances tests. First, about the hard drives I will use. The 2 drives are both 2TB but one is a 5900 and the other one is the 7200. I' using here 2 different drives to see how performances from one can impact the overall system based on the architecture used. The first one is a Seagate 2TB 5900RPM/64MB/SATA-6G and the second one is a Seagate Barracuda 2TB 7200RPM/4MB/SATA-6G. I will test the performances using RAID0, RAID1, LVM, JBOD and raw drives configurations. Mother board only have SATA2 capability. System is installed with a Debian Wheezy distribution on a 128G SSD drive. Performances will be measured using Bonnie++. Both read and write tests.
As you can see, JBOD and LVM are very close. While RAID0 is way faster. I will let you take a look at the results and decide what's the best for your needs.