steeleclipse said:
I was going to set up a RAID (striping) configuration on a DP 867 MDD Powermac (512 RAM), and was wondering if it was a) stable, and b) that much faster. Also, if I was to get a SATA controller, and use 10k RPM Raptor drives in RAID would I get the full potential of these drives?
I have a feeling the SATA controller card would be the bottleneck. If my assumption is correct, I am thinking 2 7200 RPM 8 meg cache drives using IDE RAID would work well... what do you think?
Bottom line, will I notice that much of a difference performance wise adding RAID to my system?
Thanks in advance
http://www.barefeats.com has run a number of benchmark tests on various RAID configurations (often with FireWire drives though), so you should take a look over there to see some performance figures. As for the SATA card, it shouldn't be a bottleneck. Assuming that the controller card has at least two SATA channels (so you connect each drive to its own SATA connector), then each drive has 150MB/sec of bandwidth available to it. This is much more than any single drive is capable of physically delivering (except for maybe when you're bursting data straight out of the 8MB on-drive cache). It depends on how much other traffic you have on your PCI bus also as to whether the card will be a bottleneck, but I doubt a pair of drives is going to be a serious problem for a regular PCI bus. If you're really serious about high-performance disk arrays, you will need to look at 64-bit PCI or PCI-X cards.
As JFreak as already mentioned, RAID-0, or striping, is only of benefit if you a) don't care about your data or b) back it up regularly. Overall reliability
decreases as you add more drives to a RAID-0 system, as a single drive failure in the array will kill the entire RAID set.
RAID-1 (mirroring) is safer (but more expensive, since you lose 50% of your drive capacity) because of the mirroring. It also increases read performance, as RAID-1 algorithms can have both drives in a mirrored-set reading complementary portions of the same file, increasing overall speed. I think overall write-speed may be the same, or slightly decreased compared to a single drive though.
Again, as JFreak has mentioned, RAID-5 has a kind of mix of attributes of RAID-0 and RAID-1. A RAID-5 drive set has a portion of each drives space reserved for parity information, which can be used to rebuild the contents of a drive should it fail. You generally need N+1 drives in a RAID-5 configuration to get equivalent storage to N drives, because of the storage overhead of the parity information. A lot of serious RAID-5 systems also contain at least one hot-spare drive, which the RAID controller (or software algorithm) can use to rebuild the contents of a failed disk onto in the background (using the parity information) in the event of a single disk failure. Read performance is improved, since there are multiple drives able to be accessed in a complementary fashion to read a requested file, but write performance can be degraded in 'low-end' configurations as their is a computational overhead in calculating and writing the necessary parity information. Hardware RAID implementations get around this write overhead by using dedicated RAID coprocessors to handle the computational overhead, and also employ large RAM caches to buffer reads and writes. An IBM 'Shark' storage system that we had at work a while back had 2GB of onboard cache for the drive controller! Mind you, the storage cabinet was about the same size as the 16-way p690 server it was attached to and possibly more expensive, so you get what you pay for
Finally, for those wanting fast performance as well as protection, you can combine RAID-0 and RAID-1 together to give RAID-10. RAID-10 is built by mirroring (RAID-1) striped sets (RAID-0), but it's expensive due to the high number of disks required per unit of storage.
Ummm...so...basically JFreak did a better job of explaining that in fewer words
, but I don't think an SATA PCI card will be a big bottleneck, so go ahead and experiment with some drives and let us know how you go! Remember to check the barefeats benchmarks to see if all the messing about is actually worth it though. The types of file access that you perform may not be the types that RAID sets actually speed up...
