1

I'm well aware of the widespread recommendation that any disk system that is holding a database should not be RAID-5 because of the poor write performance on RAID-5. You can look at BAARF's website to see the arguments over RAID-5 and the problems with it for databases.

I'm building an IDOL indexing server - our first - and I was wondering if people who have experience with IDOL know what the read/write balance is. RAID-5 is faster for read-heavy systems; RAID-10 for write-heavy systems (if more than about 5-10% of your disk accesses are writes, you're better off with RAID-10). I would expect that IDOL does enough writes to justify RAID-10, but I don't actually know and I was hoping someone else here does.

... so the question is: Is an IDOL indexing server likely to perform enough writes to justify RAID-10 over RAID-5?

Improve this question

4 Answers 4

Reset to default
1

RAID-10 is the preferred option in most circumstances, particularly if you are using large disks.

RAID-5/6 leaves you:

  • More vulnerable to data loss in certain circumstances
  • A long period of compromised performance in the event of disk failure
  • Less redundancy than RAID-10
  • Significantly more free disk space than RAID-10
  • Less architectural flexibility than RAID-10

Additionally, RAID-5 is not faster RAID-10 for most read purposes.

Edit: Changed marginally more disk space to significantly more disk space.

Improve this answer
4
  • 2
    Marginally more free disk space than RAID10? That is not correct at all, except two-disk RAID1 (not 10) vs 3disk RAID5.
    – pauska
    Jul 20, 2009 at 13:14
  • 1
    Actually RAID-6 is more reliable than RAID-10. You can lose any 2 drives, while RAID-10 allows you to lose one drive for each pair. Unfortunately rebuilding a drive stresses the remaining member and is particularly dangerous in RAID-10, while rebuilding a failed drive in a RAID-6 stresses equally all remaining drives and keep you protected in case of a second failure.
    – wazoox
    Jul 20, 2009 at 13:32
  • RAID-5 will have more free disk than RAID-10. If you have, say, eight disks, then your useable space is 4*disk size in RAID-10, but 7*disk size in RAID-5.
    – Richard Gadsden
    Jul 20, 2009 at 13:46
  • RAID is not backup. The whole point to is to provide redundancy and keep the service up. Rebuilding a mirror is just moving blocks around -- a much quicker option than does not completely compromise your application. I don't worry about stress out disks, I worry about stressed out users. A long-duration outage for a RAID-6 rebuild is going to stress them out.
    – duffbeer703
    Jul 20, 2009 at 13:59
2

I do not know about IDOL specifically, but during day to day operations I would not expect an indexing service to generate many write requests unless the content being indexed changes a lot regularly.

The initial indexing of blocks of new content will produce a lot of writing though, as will any re-index operations that are started for any reason. For either of these operations you will notice a significant performance difference between RAID10 and RAID5/6.

The other operation that will produce write is logging, but is is only going to be an issue if the service is configured to record copious amounts of information as the index is accessed and if that is the case then you are likely to have your logs on another volume anyway.

Improve this answer
1
  • There will be about 20,000 new files being indexed per day - the initial load is about 10,000,000, just to give you a sense of the system scale.
    – Richard Gadsden
    Jul 20, 2009 at 13:51
1

I agree with the other posters about the particular details around RAID 5/6/10 and such so I won't add much on those points. But you didn't specify what kind of hardware was being used, so I remind you that using a solid hardware RAID controller should also be kept in mind.

I've done software RAID before and after several failures, I've sworn them off. After using hardware RAID (such as Adaptec and 3Ware), the difference is substantial. Not only do they perform better then software RAID but I have yet to run into a failure with any disks. I'm not saying it will never happen, but hardware while being more costly will definitely help you with performance and availability. Don't neglect the RAID controller!

Improve this answer
1
  • Good advice - I'm going to use a DELL PERC controller, which I think is a rebadged LSI MegaRAID card.
    – Richard Gadsden
    Jul 22, 2009 at 10:05
0

Many setups include two arrays; one RAID-5 for data and one RAID-1 (or 10) for logs. I've never worked with any database wich required RAID10 for data storage, but maybe others have..

Improve this answer
1
  • Using RAID 10 here for the storage. Logs are not important - sorry. Logs are continous write, a LOT less stress than what goes on on my data.
    – TomTom
    Mar 9, 2010 at 20:59

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .