The average write size from a database can dramatically alter the performance of a RAID-0 array.
I have an application where I write documents of between 1.5-3.0 KiB. I insert about 2,500 of them a second into my test rig. And end up with about 100 million documents during the run.
The single drives I have been using cannot keep up with the load. The I/O channel is only about 10% utilized. I am putting a pair of RAID-0 arrays into service, one each for data and indices.
The big question is, of course, what size stripe should I use for Couchbase? I doubt it is writing a single document followed by its durability block. I have a wide range to choose from. The larger the block, the faster any RAID-0 system will go.
Any thoughts? Advice?