Marc ‘Driving without Due Care and Attention’ Farley has posted his top 10 Storage Innovations on his blog and unsuprisingly Thin Provisioning makes his list but he also gives a passing mention to wide-striping and suggests that might turn out to be as innovative. And he’s right; in fact I would suggest that it is more important and will be foundational for all new arrays based on spinning rust (apart from the Platypus using NetApp, which stripes wide but in a diferent way).
Now whether 3Par can claim the innovation or not, I’ll leave to our current community historian, Stephen Foskett but from an end-user point of view it is massively important. Our storage estates are simply too large to manage at the micro-level generally, wide-striping allows us to manage at a macro-level.
The next big thing is policy-driven storage management; no, not Atmos; but assigning policies to Luns defining performance, protection, ILM-flows etc. Yes, it already exists today but in my experience, it’s not widely used and trusted. Yes, you could do this without wide-striping but I think it makes the whole thing easier from an implementation point of view and hey, it might discourage some my admins thinking that they can do a better job than the automated routines…sometimes they can but life isn’t long enough.
Wide striping is a side effect of better storage virtualisation imho. Only when that’s more advanced (and there are a number of missing components in everyone’s technology right now, see Tony Asaro’s partial list on Marc’s blog http://www.storagerap.com/2008/11/top-10-storage-innovations.html#comment-139434074) will business policy management make sense.
Right now, policy is often “Do A except if…” and the exceptions are the bit that’s difficult to manage. And policy A is very often not a business policy, but a technology policy. Kind of defeats the purpose when they’re difficult to match up one to one.
Is it important to match up one-to-one? Not sure that it is.
Interesting you should mention how admins believe they can do a better job than a system that automates wide striping. As you point out, they will either a) fail, or b) waste far too much time in exchange for a negligible advantage.
Wide striping always wins. In fact, we discovered at 3PAR that our engineers could not manually tune the system to be as fast as the system’s automated wide striping.
We let customers turn all the knobs if they want to because lots of them want to – but it probably doesn’t yield much, if any, advantage. In general, they are better off spending their time ensuring that system resources are balanced so that wide striping can be as effective as possible.
Martin; no, it’s not essential to map one to one. But the engineer in me likes to see one knob connected to one dial. IMHO we have far too many knobs to twiddle and not enough dials, and far too few knobs and dials at any level higher than the technology we’re trying to manage.
I agree that the business rules are generally not well-defined (if at all); generally we get requests along the line of ‘we want some storage with some performance, we are not sure what…’. That is a hard problem to fix unfortunately.