Single Drive vs. 2- drive RAID 0 vs. 3-drive RAID 5

[EddNog]

I am confused regarding RAID 5 performance after reading the guide at Storage Review.com and reading a RAID card comparison at Anandtech so I will ask the highly knowledgeable group of people who browse this site. Right now I have a FastTrak100 card, am using 2 WD 400BBs in a striping configuration with stripe size of 64kB. Now, I am considering upgrading to a FastTrak SX4000 with 256MB of cache, buying a 3rd 400BB, and switching over to a RAID 5 array with 3 400BBs with 128kB stripe size. How would a 3-drive RAID 5 array compare to a 2-drive stripe array in terms of:
1) Random Read performance
2) Random Write performance
3) Straight Read performance
4) Straight Write performance
Please keep in mind: the 256MB of cache, the fact that I work with a large(r) than normal amount of large(r) files, and that I defrag EVERY NIGHT.

-Ed

 

[Woodie]

Assuming that this controller will do RAID 5 on IDE (I have little knowledge there, I use all SCSI raid)...

RAID 5 usually performs slower than either RAID 0 or 1, because of the parity calculations. Since you're talking just about IDE, I think the prior statement would hold true.

IF you were switching to a SCSI RAID, then it would be a toss-up: RAID 5 requires more calculations, but the SCSI controller does it instead of the CPU (which is used in IDE RAID).

 

[EddNog]

The Promise FastTrak SX4000 is a hardware RAID controller with integrated i960 processor to perform the RAID 5 parity calculations. This is also why it takes cache memory from 64MB up to 256MB, and it even supports 66MHz, 32-bit PCI bus in the case where should I purchase a motherboard with higher performance PCI bus, it relieves the bottleneck of the slow 33Mz, 32-bit PCI bus that is common to standard motherboards.

-Ed

 

[EddNog]

Actually, let's say that instead of RAID 5, I decide to spendthe extra money and buy the 4th 400BB. I would build a RAID 0+1/10 array. How would the new array (quad 400BBs, each on independent ATA/100 channel, 2 striped, 2 mirrored), on the new controller (Promise FastTrak SX4000 HARDWARE RAID controller with 256MB cache, support for 4 INDEPENDENT, single-drive channels and either 33MHz or 66Mz, 32-bit PCI bus[133MB/s or 266MB/s bandwidth]), compare to my current array (dual 400BBs, each on independent ATA/00 channel) on my current controller (Promise FastTrak100 SOFTWARE RAID controller) in terms of:
A) Random Read performance,
B) Straight Read performance,
C) Random Write performance and finally
D) Straight Write performance?

-Ed

 

[thomsbrain]

i would imagine all that cache and added PCI performance (assuming your mobo supports it) would add a lot to the burst performance, and help a little with sustained transfers when you're doing reads and writes simulataneously. it would help a lot with writes, too, they'd basically be instantaneous up to 256 MB, plus while it was filling the buffer, the drives would be writting, so it would last even longer. so that would rock. but basic read performance won't change too much, since it would be drive limited.

 

[Agamar]

I get all my SCSI raid with at least 32M of ram...I have one that has a regular P-100 chip on it..Kind of wierd to think about it.