Cost efficiency: Raid 51 vs Raid 6

[qingdom]

Hi,

I'm in the mist of planning a new raid for my farm. I'm familiar with Raid 0, 1, and 3's, but have no experience whatsoever with Raid 5's and 6's.

I've built combinations of Raid 01's, 10's, and 31's before but have never peiced together a 5, 50 nor a 6.

{EDIT}:Has a raid 51 solution been put together before?{/EDIT}

My question is, which (between 51 and 6) is more cost efficient to put together and which would have more troubleshooting headaches?

Raid 51 would require more drives to piece together, but should two drives fail in either mirror, I could theoretically replace both drives without having to reduplicate the entire array, right? theoretically (o_o)

Raid 6's requires a card with XOR with a separate processor/ram on card. Initial cost would be an arm and a leg though.

Suggestions? Recommendations?


On a side note, is there software raid 6 solution available? I have several PCI single board pc's lying around with PPGA celeron's and 64mb pc133's if it could be given a second life.

 

[fishmahn]

I think this will answer your question about RAID 6.

From my brief glance at it and your question, 6 will require less drives, and will give you 2-drive redundancy (any 2 drives can go down). 50 will require 6 drives and you'll only be able to loose 1 of them. 50 will be much faster than 6.

Hmmm, I read again and you mention mirrors in 50. 50 doesn't mirror, it stripes... 51 does mirror (5+1). Am I confused or did I make you confused?

Mike.

 

[qingdom]

You're right, mike. I need to edit my post.

I was actually thinking of Raid 51's and totally misinterpreting Raid 50. Some cards are listed supporting 0/1/0+1/5/50/6, and I misunderstood 50 as a mirror instead of a stripe.


Raid 5, i've read, needs at least 3 drives to be ideal, and is much like Raid 3 except the parity isn't stored on 1 dedicated drive, but spanned across all drives.

so a Raid 51 needs (n*2) drives to build, in the least case, 6 drives (3 + 3 [mirror])

Raid 6 needs at least 4 drives to be ideal, i calculate. it needs at least one drives (maybe more) worth of space to hold 2 parity bits spanned across drives. Hard drives are cheap. But the card itself is expensive.


I was wondering if anyone can point out the pros and cons how troublesome it would be to troubleshoot failures of maintaining a raid 51 and 6.

I got the cost factor down pretty much between the two, unless someone can tell me that raid 51 is not doable at all. =

on paper it looks possible, but has raid 51 ever been done?

 

[linux_0]

RAID is great but CAN become a huge headache when it breaks.

I have had to rebuild degraded RAID arrays before with high-end hardware RAID controllers and depending on the size of the arrays, can take 8-48 hours to rebuild.

Use a RAID controller with battery backup whenever possible (BBU) otherwise if there is a power failure anything in your write cache will go buh bye.

I may be wrong about this but I do not believe RAID 6 is possible or desirable in software.

 

[RichPLS]

I havent seen anyone, even posting here, seems more like a corporate server type to me, where money doesnt matter... I mean, 6 drives min and capacity of only 4???

 

[linux_0]

Here are some IOZONE benchmarks I ran.

https://66.235.243.163/bench/index.html

FC3 i386 is on the left FC3 x86_64 on the right.

FC3 x86_64 destroys i386 on a Dual Opteron 252 with 4GB RAM and a 3ware 9500S-8 in RAID5.

RAID6 seems to make more sense than RAID5 since you can afford to lose 2 hard drives instead of just 1 in RAID5, however RAID6 is obviously more expensive.

RAID51 is more expensive still.

What kind of servers and controllers are you considering purchasing or do you currently have?

Good luck!

 

[fishmahn]

Ok, here's my take on why someone would want RAID because as I see it, it only does 2 things (these are the desired results, not side-effects like reduced performance due to parity calculations) results:

1) increase performance
2) provide redundancy to prevent unscheduled downtime.
RAID does not provide a backup solution so that's not a benefit, it only prevents unscheduled downtime.

RAID 0 increases performance.
RAID 1 can increase read performance (most won't notice) and provides redundancy.
RAID2,3,5 can increase performance (5+drives in HW RAID5 to see it) and provides redundancy at a lower storage 'cost' and higher driver requirements.
RAID 10 (1+0, 0+1, whatever, it's almost the same) gives you a cheaper way to get both increased performance and redundancy on a cheaper driver requirement.
Others: 51, 50, 6, 60, etc., all try to further increase the redundancy so that more drives can go bust at once without it being catastrophic.

If money is no object, then speed and redundancy are the only concerns. If speed is the main concern, anything with a 0 in it will be fine, RAID 0 being the best.

If multiple failure is the concern, then RAID 6 is going to be best because it can support any 2 drives going down in the array - that's 50% failure in a 4-drive array. It has the same 'wasted' space as mirroring, but in a 4-drive mirrored array, if both the master and mirror copy of "C:" drive goes down... No other RAID you've mentioned so far has the same level of failure protection.

I'm willing to bet there's another RAID (7 anyone?) that expands on 6 and can create 3- or 4-drive redundancy, though I think that may be a little extreem except in unmanned situations (space, undersea, etc.)

Mike.

 

[SB02]

Hello,
Raid 7 (NON STANDARD, as in not official I just seen it)

Storage Computer Corporation has RAID 7, which adds caching to RAID 3 and RAID 4 to improve I/O performance.

I seen it online, don't know a thing about it, just thought I’d throw it out there

again from a website not of my own I found:

RAID 1+5 and 5+1 might be sarcastically called "the RAID levels for the truly paranoid"

So if 6 is so much better then 5
What could 6+1 support for number of drives failing
Let’s say you have 2 servers with 4 drives each
The RAID 5 example
5+1 can lose up to 5 Drives if at least 1 of the 2 servers has no more than one failure!
I.E.
Server 1: Drive 1 FAIL, Drive 2 FAIL, Drive 3 FAIL, Drive 4 FAIL.
Server 2: Drive 1 FAIL, Drive 2 Good, Drive 3 Good, Drive 4 Good.
5 of 8 is pretty good
Now the RAID6 example
6+1 can lose up to 6 Drives if at least 1 of the 2 servers has no more than two failures!
I.E.
Server 1: Drive 1 FAIL, Drive 2 FAIL, Drive 3 FAIL, Drive 4 FAIL.
Server 2: Drive 1 FAIL, Drive 2 FAIL, Drive 3 Good, Drive 4 Good.
6 of 8 is also pretty good
Might not be worth the expense but hay some of us are “truly paranoid”

just my thoughts (please don't hate me)
SB