4 Replies Latest reply on Jun 18, 2009 12:04 PM by Ndi

    RAID 10 performance

    Ndi

      Hi, I'm on an ICH 9R, running on four (wish it could do six) drives. There's been some issues with the drives and they are failing.

       

      As they start to go, I am upgrading to another manufacturer and better drives. These drives are also faster, so I was wondering in what order should I do the replacement. It is likely I'll run a while with 2 old and 2 new drives (same capacity).

       

      Since it's a stripe of mirrors, should I go and

       

      a) replace 1 and 2, speed up one branch (and rebuilds should the case arise) or should I go for

      b) 2 and 4 (or 1 and 3)?

       

      In theory, the fastest drive will reply first on read and only slow for the slower drive on long writes. So on read I should get a boost from the new drives, and get (almost) nothing from the writes.However, if it's not fastest-drive-first, then speeding up a branch will still get half the advantage from the striping both reading and writing, PLUS faster rebuilds on the upgraded side.

       

      How does that work in practice?

       

      Thanks.

        • 1. Re: RAID 10 performance
          UglyPercy

          If you're running IMSM under Windows 7, I do not believe that your drives are actually failing.  I believe IMSM 8.9.0.1015 is not working correctly under Windows 7 RC1, and that it incorrectly declaring that you are experiencing hardware issues.

          • 2. Re: RAID 10 performance
            Ndi

            Well, thanks, but the drives that failed have been plugged in different PCs and they show SMART failures, bad sectors, and at least one can't be formatted at all.

             

            However, that wasn't my question, I was wondering about performance, and would it be better to rebuild with one drive on each branch or rebuild with both drives on the same branch.

            • 3. Re: RAID 10 performance
              UglyPercy

              Sorry, didn't quite understand that.  If different PC's are reporting failures on the drives, that is a very strong indication that the drives are bad.  I'm just not used to the idea of multiple modern hard drives failing independently at pretty much the same time.  Sounds like a major manufacturing defect to me.

              • 4. Re: RAID 10 performance
                Ndi

                Yes, well, that wasn't the question. There's another thread you replied in, about port 4 failure. This is about RAID performance.

                 

                Anyway, on-topic, I upgraded one drive today. The controller was mean to me. First, when I added the drive and tried to rebuild to this drive, IMSC claimed that "Wizard could not start the rebuild process". This was odd, especially since the right-click menu read "reset this drive to non-raid". However, after rebooting I could finally add it (On boot the BIOS reported 1-2-3(RAID)-4(nonRAID). Then it rebuilt to the drive, but showed 5 drive points in the tree. 1, 2, 3, 4(rebuilding) and another one that read "missing hard drive". Oh well, after rebuild it figured itself out. Great.

                 

                Anyhow, I ran a large HDTach on it and I get the same result as before to the decimal. So one faster drive sped nothing up.

                 

                Since it's a RAID10, the first stripe member (1+2) were untouched so any increase in member 2 (3+4) would have showed. As a result, I think the second mirror set (3+4) had the exact same performance as the old one. This means that not only the fastest drive doesn't get heads-up, the controller waits for the second drive's opinion before serving the data. Either than or WD lie on their specs sheet

                 

                I kid, even if the drive was slightly slower, it had to outshine the old drive SOMEWHERE on the graph. The graphs were identical to the point where you can't see a red dot (the blue (old) sample covers it). So it's identical. The only explanation I see is that the fast drive waits for the slow drive. So my answer is clear (though untested): I need to upgrade one member entirely (1+2 or 3+4) before any gains in performance. This is nice to know.