Systems Engineering and RDBMS

RAID Levels and IOs per disk

Posted by decipherinfosys on March 1, 2007

Here is the calculation of the I/O’s per disk for different RAID levels:

  • Raid 0 — I/Os per disk = (reads + writes) / number of disks
  • Raid 1 — I/Os per disk = [reads + (2 * writes)] / 2
  • Raid 5 — I/Os per disk = [reads + (4 * writes)] / number of disks
  • Raid 10 — I/Os per disk = [reads + (2 * writes)] / number of disks

So, as you can see from above RAID 5 incurs a higher overhead for writes as compared to RAID 10 which is why for requirements of a highly performant and transactional RDBMS, it is rarely recommended to have RAID 5 for the logs.

Also, how can you make use of this information to see whether you are running into a I/O bottleneck or not – because, what is presented above is just theory and to prove a well known point about RAID 5 vs RAID 10. Well, if we take the Windows OS for example and use perfmon (performance monitor) utility to measure these counters:

PhysicalDisk Object: Disk Reads/Sec., Disk Writes/Sec., Avg. Disk Queue Length (you can descriptions on these by hitting the explain button in perfmon when selecting the counters). Now, assume that you have put your log files on RAID 1 system which has 2 physical disks on it. Since logs are written to sequentially, RAID 1 is a good choice. Suppose, the counters measurements over a period of time yield you these values:

Disk Reads/sec = 90
Disk Writes/sec = 80
Avg. Disk Queue Length = 5

In that case, you are encountering (90 + (2 * 80))/2 = 125 I/Os per disk and your disk queue length = 5/2 = 2.5 which indicates a border line I/O bottleneck (any value over 2 is a cause of concern and should be evaluated along with the I/Os seen per disk).

From a RDBMS perspective, all three leading RDBMS: Oracle, SQL Server and DB2 LUW provide with system level information to look into the wait events for further diagnosis. You can look at the pending I/O requests, the latch waits etc. but do not (of course) provide any visibility into the physical disks experiencing the problem. In a future whitepaper on our site, we will cover in detail the different wait events in the three leading RDBMS and how they can help you to troubleshoot performance issues.

Sorry, the comment form is closed at this time.

%d bloggers like this: