I missed this in my regular trawl for new storage blogs! A blog authored by Moshe and the guys at XIV.
And although I’m sure the guys are more intelligent than me but I just don’t get what they are trying to say. Well, I do but the blog appears to be smoke and mirrors at the moment.
From the blog
Here are our final conclusions:
- 7200RPM drives are much more cost efficient, energy saving, and dense.
- There is no real problem in satisfying overall business needs with 7200RPM drives, assuming the proper architecture is in place.
- We need a storage architecture that will enable us to really exploit all drives equally, without manual tuning. We can’t afford having 80% of our transactions handled by only 20% of our disk drives.
Well duh to points 1 & 3; well certainly cost with regards to capacity but I don’t get point 2 even trying to follow the logic. Lets say I have an application suite which peaks at 32,000 IOPs and this is alot of small block random read I/O, so not especially cache friendly; so I’m going to need about 320 7200RPM spindles (and I’m excluding any overhead for RAID of any kind and I am going to assume that I can evenly spread my load across spindles).
However, my application only needs about 26 Terabytes of disk; I am looking at wasting a huge amount of available capacity on the larger SATA drives, what I really need is fast, smaller disks. Flash is still a little bit expensive for us at the moment.
Pity my workload won’t even fit on an XIV at the moment either
Or perhaps the key word is ‘satisfying overall business needs’ and that XIV will fulfill 80% of my requirements but so would Clariion/NetApp/DS4800 etc. This feels very much like a Generation 1 product, call us in 18 months please.
I’ll keep following the blog tho’; it’ll make me smile!!!
“Lets say I have an application suite which peaks at 32,000 IOPs and this is alot of small block random read I/O, so not especially cache friendly; so I’m going to need about 320 7200RPM spindles”
Okay. I’ll play along. Let’s say the application needs 150,000 random IOs. We highly recommend flash drives and lots of them in a DMX. Congratulations, sign here.
The whole point of the architecture is you are spreading the LUN across 180 disks. Very easy and clean interface. Forget LUN layouts, it’s in there. Does it meet every need? Of course not. Will it meet the vast majority of needs? Certainly.
I’ve been in a number of shops that think they are doing tons of IO. Mostly, it isn’t the case when you drill down on their frames.
I reached the same conclusion, in my opinion they’re approaching the right problem with a wrong architecture.
And they seem to have completely forgotten the green factor.
Actually Rob, I was taking a real world example. I know what the I/O rates are on my key applications.
I guess my real problem with XIV is that it really doesn’t scale for my needs at the moment and I’m struggling to see how it will.
“I guess my real problem with XIV is that it really doesn’t scale for my needs at the moment and I’m struggling to see how it will.”
Sounds like it won’t. Just because you have a hammer in your hand, not everything you see is a nail.
Okay, the problem is that at the moment; IBM are positioning the XIV array fairly and squarely at the enterprise market. And at the moment, it neither scales capacity wise or performance wise; I think it’s going to need both to get any kind of market traction.
I don’t think there’s any reason why it won’t scale; for instance, if you put an Infiniband ‘backplane’ it wouldn’t half look like an Isilon box and it might make an interesting bulk storage device. There’s going to be loads of spare CPU cycles in the box; perhaps run Diligent in it. You’ve got an interesting deduped back-up to disk device. But a Tier-1 Enterprise array, not yet.
And oh yes, the XIV GUI is gorgeous! I have to say it’s the best looking Storage Management GUI out there.
I’m a big fan of the 3Par GUI and I must say that the XIV GUI looks better!