0

I've set up FreeBSD 7.X using gstripe (RAID 0) with two 320 GB SATA disks and raw read speed went from 70 MB/s to 110 MB/s. I'd like to know if I can reasonably expect similar gains (say roughly 200 MB/s) if I add two more identical disks to the stripe set (I understand I may lose my current data in the process of reinitializing the stripe, that's not a problem right now.) A related question would be how many disks can I put in a stripe set before the speed gain flattens out?

I've got an Intel Q9550 Quad CPU and 8 GB RAM so I don't think that's going to be a bottleneck. Using ASUS motherboard SATA ports. I'm currently reading about 500 GB multiple times more or less "sequentially" as part of a data-mining experiment. I'd like to save the $$$ and avoid the headache if this is going to be a dead end. TIA

1 Answer 1

2

Yes, in terms of performance you can expect to get just about the sum of all the drives. Do note that your seek time will be the maximum of all drives. The slowest drive will determine your seek time.

Your speed gain will be limited by the rest of the computer. Most likely your SATA controller.

5
  • Is this theory or are you speaking from experience? How many drives do you have in your biggest stripe set? Why do you say slowest drive determines seek time? Seeks are overlapped with data transfer so that if you have enough drives the contribution of seek time to read delay should approach zero. CPU should see a steady stream of full buffers to read if the stripe is working at max efficiency. I'm trying to find what the "magic number" is for number of drives to reach that ideal.
    – hotei
    Jul 31, 2010 at 1:29
  • My SATA controllers are rated at 300 MB/s. I have 4 of them that work independently (or so I'm told). At present both of them are working at 70 MB/s. If I have 4 disks then each of them will still be working independently at 70 MB/s. I'm pretty sure the motherboard data paths are good for several GB/s so I don't think that's going to limit me either.
    – hotei
    Jul 31, 2010 at 1:48
  • @Hotel: In my workstation I have 2 raid sets (both raid 6) one with 16 drives and one with 8 drives in an external enclosure connected by minisas. The only bottleneck for me is the raid card which is limited by it's cpu at around 700 MiB/s. As for the seek times, all the data is stored only once, so when seeking for some bit of data the slowest drive will determine the overall speed. The only "magic number" is the amount of drives needed to saturate one of your buses.
    – Wolph
    Aug 1, 2010 at 1:50
  • When you have 4 drives your bottleneck will most likely still be the drives. After that it will probably shift to the pciE bus that your controller is connected to. If you'll add more controllers you can increase that again. Unless you have a dedicated RAID controller your bottleneck will most likely be the bus of your controllers (or the drives if you don't have too many).
    – Wolph
    Aug 1, 2010 at 1:53
  • Ok, sounds like it should be worthwhile to load up with another two drives at least. My requirement is 95%+ sequential reading so seek shouldn't be a problem.
    – hotei
    Aug 1, 2010 at 3:10

You must log in to answer this question.

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