It’s that time of year where vendors seem to all want to come and see us at the moment; ‘new’ products to talk about, discovering areas where they can ‘help’, any excuse to try and find out what budget I might have going forward (very low BTW, I’ve spent it all with someone that you aren’t).
And at times they come out with some absolute gems and I had one today; a nameless vendor came in today and tried to convince us that their storage was much more space-efficient than anyone else as they had no hardware RAID overhead. Really, I ask?
Oh yes, we split files into chunks and distribute those chunks with parity data across the disks using software.
So a bit like RAID really but done in software?
I suppose you could say that but we get 83% of the disk available to allocation.
Really, so a bit like RAID5 7+1 or RAID6 14+2 if you exclude the overhead for hot-spares etc. Clever, you’ve invented software RAID.
BZZZZTTTTT…Next please!
Okay a bit harsh but really I wasn’t born yesterday and if you are going to protect my data with some kind of parity scheme, you are not going to be more efficient than RAID-4/5 and probably not more efficient than RAID-6/DP with large SATA drives.
P.S Thanks to all of you who have read and even commented on the blog, I’m now at the stage that I have to pay Typepad and if no-one had come, I might have decided to can it. But with over a 1000 hits, I’m quite chuffed and will definately carry on.
you know, you can increase your storage on TypePad too. It’s fall, you can always use more blog. 😉
joking aside, we’re glad you’re sticking with TypePad.
Then there’s the problem of assesing risk with some of the more esoteric schemes on offer.
Take IBM’s XIV and the implementation they use as an example.
What RAID is that? I’ve heard it described as RAID10; by someone who should know better too. It’s not a disk mirror like RAID10, it’s a block mirroring system; and it stripes across all 180 1TB drives to boot. Unfortunately, if a block fails to read or write, the chances are the disk is close to death, if not dead already. The unit of weakness is the disk.
Now, get a block read failure while doing the needed rebuild on any one of the remaining 179 disks and you lose all your data. That’s not acceptable. It’s a software version of RAID5 at 179+1, which is crazy.
With RAID, regardless of what nummbers or letters are tacked on the back of it, it’s essential to ask; what’s the chance of some problem during rebuild? Can my data survive?
If the answer is no, in my book, it’s not RAID. That R shouldn’t stand for Ruinous.
Good to see you are staying. At last I’m no longer the ‘new kid on the block’ – hehe
Seriously though, good to have an independent view from someone who has to put up with using all this stuff day to day.
@Alex
“What RAID is that? I’ve heard it described as RAID10;”
They call it RAID-X. 1MB blobs that have their associated copy on a random disk (not on same shelf 😉
“Now, get a block read failure while doing the needed rebuild on any one of the remaining 179 disks and you lose all your data. That’s not acceptable. It’s a software version of RAID5 at 179+1, which is crazy.”
Yes – software read on rebuild is bad.
I’ve worked with REALLY important data before. Not only was it mirrored at the frame, it was software mirrored to another frame in another data center. With XIV you can buy two for the cost of some systems and do likewise (if it suits your fancy).
Hi,
Its so nice to see a user blogging about this stuff…. I was getting so sick of all the vendors taking pot shots at each other…
🙂
@Rob
Yes, replication is always a requirement, regardless of RAID. But having to do a complete datacentre DR because you’ve just had a couple of disks fail in your RAID-(nowhere near as good as dual parity) storage sounds painful — and much more expensive!
BTW, NetApp uses software RAID within their appliances, not hardware based RAID, not that is matters. Also, glad we have an end-user keeping us all honest! It is hard to appear unbiased when Dell pays my salary, however, they don’t host or pay for my blog (was there before Dell, and will be there after Dell).
My problem is…I’m passionate about certain technologies, one of those being PS-series storage! Keep up the writing!
Well, you could probably make an argument that all RAID is implemented in software at some form or another.
And bias is fine as long as it is declared; I don’t read vendor blogs to look for unbiased opinions but I read a variety to get various opinions. Even as an end-user I have biases; some are rational, some aren’t!
I worked pre-sales for IBM reseller a while ago; I think that I was one of the first non-IBM employees in the UK trained on ESS. I also think that AIX/pSeries (I was a Certified Aix Technical Expert) is the best Unix.
I run a SAN with NetApp/DMX/CX/DS48k in at the moment. And I’ve been a reference for both IBM and EMC in the past.
And I carry my own baggage.
“Well, you could probably make an argument that all RAID is implemented in software at some form or another. ”
No. In fact, LSI/IBM DS series has custom ASICs to perform the parity calculations. That is why IBM was late to the RAID6 party. I’ve heard on more than one occasion (so it is either right, or consistent 😉 that the RAID6 overhead versus RAID5 is 3%. Other vendors do much worse.