0

I am aware that RAID comes with certain write penalties, etc and that varies at different levels. I am trying to figure out how latency is affected in RAID 1. From my SAN, I can look at the number of writes, reads, bytes, and latency in individual HDD. While others match, latency slightly varies. I am trying to figure out the following.

RAID 1 HDD 1 Write Latency: 24ms HDD 2 Write Latency: 28ms

Does that mean the total write latency = 24ms + 28ms? Or just 28ms? or (24+28/2)ms. I am bound to believe it is 28ms because we are trying to find the max?

sqlbuzz
  • 23
  • 4
  • What type of storage solution? – ewwhite Feb 08 '13 at 03:40
  • It is basically a SAN - MSA 2000 G3. 15k HDD, 6GB SAS switches. I know the latency is high. I am trying to get < 10 ms, so checking IO path and all kind of stuff. No benchmarking was done (I had not started working here when this was already production), so we do not have any baselines. Trying to load balance database files based on the workload. Thank you. – sqlbuzz Feb 08 '13 at 05:10

2 Answers2

4

You're completely missing the effect of the HP P2000 G3's write cache!

Random writes will hit the flash-backed DRAM write cache on each controller first, be coalesced and fed to the drives sequentially from cache.

In your setup, as long as you have write caching enabled and the proper RAID level to support your application's sequential throughput needs, there isn't much tuning you can do at the storage level. Add to that the fact that you're connected via 6Gbps SAS links!

There's plenty to start examining at the OS level, though. I/O elevators, filesystem choice, readahead, mount options, etc.

Is your application server running Linux?

Community
  • 1
ewwhite
  • 194,921
  • 91
  • 434
  • 799
1

Total write latency will be the worse of the set for raid 1. That said, it might not even be that bad. Depending on your raid device/software, it might even allow a small skew.

For example, mdadm has

--write-behind=
Specify that write-behind mode should be enabled (valid for RAID1 only).
If an argument is specified, it will set the maximum number of outstanding
writes allowed. The default value is 256. A write-intent bitmap is required
in order to use write-behind mode, and write-behind is only attempted on
drives marked as write-mostly.
R. S.
  • 1,624
  • 12
  • 19
  • Thank you. That is an excellent explanation which also verifies what I was thinking (or was more inclined towards). – sqlbuzz Feb 08 '13 at 05:08
  • This isn't software RAID... What about caching? – ewwhite Feb 08 '13 at 05:28
  • Then it entirely depends on the settings of the controller and the amount of data. The worst case is still going to be the latency of the slowest drive. – R. S. Feb 09 '13 at 19:41