I'm upgrading my wifes system, and have been looking at RAID options.
There is some amount of myth and misinformation associated with the performance benefits of RAID. Contrary to (some) popular belief, typically, single users don't benefit from RAID, since RAID systems are mostly designed to assist with queued I/O in different locales of the disk. Single user I/O patterns are typically limited in simultaneous operations and highly localized.
Storage review did a piece on this in their TCQ, RAID, SCSI, and SATA article. A priority of my upgrade project is redundant storage of data, so RAID1 and RAID0+1 are effectively my options with controller based solutions.
Another article which tested Raptor360's on a variety of on-board controllers was Chipset Serial ATA and RAID performance comparison, which included not only pure benchmarks but some "realworld" benchmarks, also.
Effectively, to get a performance benefit with redundancy, you need a 0+1 RAID solution. RAID1 even though it theoretically has the possibility for offering better read performance (through simultaneous reads on both primary and mirror drive,) these don't materialize in any of the low-end, uncached controllers, and even then, I doubt it would make much difference in a single user environment.
RAID 0+1 (on nf3) seems to show real, tangible benefit, in terms of raw HDtach avg write rates, around 35%, a penalty for HDtach avg read rates, and significant benefits in IOMeter Workstation ( even at 2 through 8 "load", although performance plateaus there) and Content Creation tests, in the order of 200% benefit.
Drive speed obviously effects these figures. The drives that I'd consider for a 0+1 array would be Seagate SATA 200GB's (or Seagate SATA 160 SATAII's, but I've seen anecdotal evidence that TCQ/NCQ actually hurts single user performance, as per storagereview's article.)
My other option would be to use a RAID1 mirror of WD Raptor 740D's, and then an additional IDE or SATA mirror, and use NTFS mountpoints within the drive to relocate Documents and Settings profiles and select Program Files directories to the larger and slower mirror, getting around the limited capacity (which is a bit of an issue.)
A 0+1 array of Raptor's is theoretically possible to budget for, but... it costs $700 or so, for 148GB of capacity, vs $500 or so for 400GB.
I'm leaning towards 4x Segate 200GB drives on an nforce3 (the DFI Lanparty UT 250gb nforce3, suggested by Hajime in reply to my initial post,) although I'm still not convinced, looking at the graphs, how a mirror of Raptors would compare to the relatively slow Barracuda 7200.7's in a real-world situation, being that the Raptor is probably 25-30% quicker to begin with, as a single drive, in raw speed.
So... if this were your $400-600 budget, which route would you go?
There is some amount of myth and misinformation associated with the performance benefits of RAID. Contrary to (some) popular belief, typically, single users don't benefit from RAID, since RAID systems are mostly designed to assist with queued I/O in different locales of the disk. Single user I/O patterns are typically limited in simultaneous operations and highly localized.
Storage review did a piece on this in their TCQ, RAID, SCSI, and SATA article. A priority of my upgrade project is redundant storage of data, so RAID1 and RAID0+1 are effectively my options with controller based solutions.
Another article which tested Raptor360's on a variety of on-board controllers was Chipset Serial ATA and RAID performance comparison, which included not only pure benchmarks but some "realworld" benchmarks, also.
Effectively, to get a performance benefit with redundancy, you need a 0+1 RAID solution. RAID1 even though it theoretically has the possibility for offering better read performance (through simultaneous reads on both primary and mirror drive,) these don't materialize in any of the low-end, uncached controllers, and even then, I doubt it would make much difference in a single user environment.
RAID 0+1 (on nf3) seems to show real, tangible benefit, in terms of raw HDtach avg write rates, around 35%, a penalty for HDtach avg read rates, and significant benefits in IOMeter Workstation ( even at 2 through 8 "load", although performance plateaus there) and Content Creation tests, in the order of 200% benefit.
Drive speed obviously effects these figures. The drives that I'd consider for a 0+1 array would be Seagate SATA 200GB's (or Seagate SATA 160 SATAII's, but I've seen anecdotal evidence that TCQ/NCQ actually hurts single user performance, as per storagereview's article.)
My other option would be to use a RAID1 mirror of WD Raptor 740D's, and then an additional IDE or SATA mirror, and use NTFS mountpoints within the drive to relocate Documents and Settings profiles and select Program Files directories to the larger and slower mirror, getting around the limited capacity (which is a bit of an issue.)
A 0+1 array of Raptor's is theoretically possible to budget for, but... it costs $700 or so, for 148GB of capacity, vs $500 or so for 400GB.
I'm leaning towards 4x Segate 200GB drives on an nforce3 (the DFI Lanparty UT 250gb nforce3, suggested by Hajime in reply to my initial post,) although I'm still not convinced, looking at the graphs, how a mirror of Raptors would compare to the relatively slow Barracuda 7200.7's in a real-world situation, being that the Raptor is probably 25-30% quicker to begin with, as a single drive, in raw speed.
So... if this were your $400-600 budget, which route would you go?
Facebook
Twitter