My question concerns only Linux software RAID ("md").

I'm going to set up a RAID1 array with three devices. I need to choose between using the third disk as a spare or as an active member of the array. There are no other arrays in the system, so there is no possibility of sharing the spare.

What's the performance impact of the third disk being routinely active? For example, what is the effect on tail latency?

If I go for three active devices, is there a data resilience advantage (e.g. can I scrub the array to take a three-way vote on the correct value of each byte? Apart from mitigating the risk of failure of a second device during reconstruction, is there any other reliability difference?

  • Why not actually setup and benchmark the two configurations, then decide? – Dan Jun 20 '15 at 19:55

Due to how Linux raid behaves, using a three way raid1 vs a two way+hot spare raid1 setup gives:

  • slightly lower random write performance, because arrays's access latency (seek time+rotational delay) is the worst of the the disks composing the array
  • significantly better random read performance, as Linux raid is capable of issuing multiple, independent reads to different disks
  • better disk failure recovery, as the third disk is always synchronized with the other two
  • no changes in data resilience, as mdraid implements no voting capability (this is by design: implementing a voting system means that each read must be issued to each disk, significantly lowering read performance)

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.