I had a chance to test HP Smart Array P800 SAS RAID controller with HP Proliant DL380 G5 server with the following configurations:

  • Dual Quad Core 2.5GhzHP DL380
  • 32-GB RAM
  • P800 Smart array with SAS interface
    • 25 disks, 10K RPM, 146G each
    • 512M cache
    • Battery backed

Here is the sysbench fileio random write performance numbers  for RAID-10 with the following configuration:

  • 18 disks, ext3 noatime, nodiratime (mirrored)
  • 0% read, 100% write cache
  • 64K strip sizep800 Array Controller style=
  • Battery enabled
  • 16K block size (sysbench)
threads sync – rndwr (reqs/sec) direct – rndwr(reqs/sec)
1 4690 4502
4 5556 5203
16 5560 5233
32 5559 5204

Overall the numbers seems to be reasonable for pure random writes with 64K strip size. The same configuration with RAID-5 drops the write performance by 1/4th. The controller does not provide a way to configure the cache burst time, and it automatically controls the cache and evenly distributes to other volumes.

Overall HP P800 might be a good choice for database workloads as it has 25 SAS disks support (depending on enclosure ) in a affordable price, even though 15K disks price to 10K price is wide; and people can simply add more 10K disks than buying fewer 15K



Related posts:

  1. Notes from Falcon from the beginning
  2. InnoDB plugin row format performance – II
  3. MySQL performance comparison
  4. Notes on InnoDB Scale on servers with many cores