Even if the Yfrog presentation on their usage of HBase has only 9 slides, you’ll find some very interesting notes in there. Here’s a couple that caught my eyes:

• 10k concurrent requests/second
• 4 HBase clusters ranging from 50TB to 1PB
• Having more smaller nodes is better than having less faster bigger nodes
• We use sub $1k desktop grade servers, they work great!^[1]
• Inexpensive 2TB Hitachi disks (~$100) work great, get more units for your money
• Load test HBase with YCBS: just leave it running for a week, if nothing crashes, you are good.

Make sure you are checking all the slides below.