Raid Confusion Please Help

[negreac]

So I'm in the process of building my main home PC and decided instead of going with my original plan of running 4 x 1TB in raid 10 or 0 to get an external enclosure/raid box since I happen to have another 4 1 TB hard drives sitting around. All 8 drives are approx. the same specs. I've listed them below:

PC Specs:

CPU: AMD Phenom II x4 955 BE
MB: MSI 790fx-gd70 (does not support USB 3.0 or Sata III)
GC: 2 x Sapphire Radeon HD 4850 (512mb) in crossfire
RAM: Currently 2 x 2GB of 1800mhz DDR3 Corsair (probably going to add another 4gb or just upgrade to 16gb)
Case: AZZA Fantom 900
SSD: Intel 160gb X25-M (for just about everything except media)
HDDs (all 1TB 32mb cache 7200rpm):
4 x Seagate Barracuda 7200.12 (ST31000528AS) ~ 130Mb/s
2 x Western Digital (WD1002FBYS) ~ 110Mb/s
1 x Seagate Barracuda ES.2 (ST31000340NS) ~ 110Mb/s
1 x Seagate SV35.5 (ST31000525SV) ~ 130Mb/s

So basically I have 2 decisions to make that I need help with.

Decision 1: To purchase an enclosure with built in RAID controller or not.

The 2 boxes I am looking at are the RJ Tech Mediasonic H8R2-SU3S2 (w/ controller) and H82-SU3S2 (w/o controller)

H82-SU3S2 (w/o controller)
H8R2-SU3S2 (w/ controller)

I am leaning toward just getting the one with the controller built in rather than using a separate PCI-e card controller or software raid simply due to the fact that it is one less thing I will have to mess with when I upgrade my computer next. (I'm assuming the box simply shows up as a single big ass external hard drive). I have been able to find very few reviews on it though and worry about the capability of the on board controller. Does any one have any thoughts on whether this seems like a good or bad way to go and for what reasons?

Decision 2: Whether to use RAID 10 or Raid 50

I completely understand RAID 10 but am still a bit hazey on how it stacks up to raid 50 and having a hard time getting straight answers. In my particular case I know the following:

Raid 10
Space: 4 TB
Read: 8x = 4(2)x (theoretically 800mb/s)
Write: 4x (theoretically 400mb/s)

Raid 50 (2 x 4TB Raid 5 Arrays striped)
Space: 6 TB
Read: 5x = 2(3-1)x (theoretically 500mb/s)
Write: Assuming there are no raid hardware limitations it should be 5 times as fast - 5x = 2(3-1)x (theoretically 500mb/s)

The most write intensive task I will be doing is backing up blu rays from 3 drives (1 at 8x-36mb/s and 2 at 12x-58Mb/s). Assuming that the write speed would end up being similar the drives' read speeds I would like to have something above 150mb/s. I know Raid 10 is capable of this but does anyone know if I can accomplish this in Raid 50? From my understanding the bottleneck would be the RAID hardware right? If so what is my best option?

I also want to be sure that RAID 50 won't present a problem down the road when it comes to potential parity errors. This is where I get a bit confused. I know once you start getting into RAID 5 arrays this large you are more than likely going to have parity write errors resulting in the inability to be able to rebuild the array. Does RAID 50 alleviate this problem? Some people seem to think it does but I don't see how that's the case since in all reality RAID 50 is no different then having multiple RAID 5 arrays. I don't see how there is any further redundancy when it comes to parity errors. The only increase in redundancy is being able to support a drive failure in each Raid 5 array simultaneously but this is completely independent of parity errors.

The intention is in a couple years to upgrade replace all of these drives with 3tb drives to make a 24tb array. Does this make a significant difference in failure potential based on what I mentioned before about parity errors? I know raid 10 is the best bet here guys but if I can have 1.5 times the storage space by using RAID 50 with minimal risk and performance loss I would really like to do that. Let me know what you think about RAID 50 in my particular situation. Thanks for all the help!

 

[negreac]

I updated my post above to try and make my questions about raid 50 redundancy and speed a little clearer as well as putting in actual theoretical speeds for a RAID 10 and RAID 50 array.

 

[negreac]

On a side note those theoretical number above are very theoretical since hardware bottlenecks will severely limit the speed. In theory raid 5 is actually faster than raid 50 but in actuality the additional parity read processes required in order to write to a large raid 5 vs. multiple smaller raid 5s (a.k.a raid 50) will require more resources from the hardware and cause a slower transfer speed. Here's a good thread explaining it further:

http://arstechnica.com/civis/viewtopic.php?f=11&t=193995

 

[negreac]

Success! I found an actual test of this enclosure with 8 3TB 7200rpm 64mb cache drives (Hitachi HDS723030ALA640 ~125 Mb/s) using RAID 50 and USB 3.0. Read was 186.7 Mb/s and write was 154.2 Mb/s. I imagine I would get similar performance based on the drive speeds.

http://www.hotway.com.tw/products/h8r2-su3s2.htm

Digging more into the parity write error (a.k.a. Raid 5 write hole) issue I am pretty certain that RAID 50 has the same issue as RAID 5. In othe rwords if you have a single parity bit written/read wrong your can kiss your entire array goodbye. I can mitigate some of the risk based on power failure write errors by using my UPS. I have yet to find out whether the built in controller has a battery back up but I doubt it based on the price point.

This doesn't eliminate the risk that I get a read error since the accepted read failure rate (a.k.a. URE) at least a couple years ago was 1 out of 10^14 bits which is 12TB. I'm hoping that with advances in hard drives lately that this has been improved but it's still a very big concern of mine especially if I ever upgrade to those 3TB drives. Even with the 6TB raid 50 I have a 50% chance of not being able to recover based on the failure rate above. Some people claim that this is still a great overestimate though based primarily on how drives fail today and new monitoring and recovery technology for both hard drives and raid controllers. You can find a couple of articles about the issue and why it might be hype here.

http://www.zdnet.com/blog/storage/why-raid-5-stops-working-in-2009/162

http://www.pizzaandcode.com/posts/780

I think I have all my answers now. I just have to double check with Mediasonic that I can move the raid from computer to computer no different then an external hard drive and find out if it contains a battery back up. If you see any problems with my logic or findings above or have anything to add please let me know.

 

[foscooter]

One thing I can say is that HDD DO NOT Scale in RAID like SSD!

A RAID 0 array with 2xSSD will double the read/writes, not the access times.

A RAID 0 array with 2xHDD WILL NOT double the read/write, only give you about a 40% boost. I have benchmarks to prove this.

I've read that RAID 10 is slow, when compared to 50.

And, if you are setting up a RAID box, via USB, you'll only get USB speeds, no matter how many drives in whatever array you have!

 

[negreac]

Based on everything I have read thus far RAID 10 should be much faster than RAID 50 if not only for the fact that there are no parity calculations. In fact overall I believe raid 10 is even faster than raid 0 when it comes to read and the same when it comes to write.

I imagine raid 10 would scale similarly to raid 0 with the same amount of stripped groups. If this is true then I would assume after about 5 stripes you are pretty much maxed out. This is solely based off an article I found here where they put 2-8 HDDs in raid 0 and tested them.

http://www.tomshardware.com/reviews/RAID-SCALING-CHARTS,1635.html

The box has USB 3.0 and eSATA. It of course is backwards compatible with USB 2.0 as well. There is no way I will saturate either eSATA (3Gb/s) or USB 3.0 (5Gb/s) so that is the least of my worries.