This is a very good piece on FAST-VP on VMAX, well-written and some good advice in it but it sums up almost everything that is wrong with VMAX today. VMAX has too many nerd-knobs and so people think they should fiddle and try and out-do the machine.
And hence probably make a right-old mess, FAST-VP ends up not working quite as well as it should and so people tend to fiddle even more and the next thing you know, you are trying to manage your VMAX in the way you would have managed an old-school Symm.
I think it is time that EMC and their users seriously consider breaking away from the past; the old-school nerd-knob fettling needs to stop. I know that is why storage admins get paid the big bucks but I do wonder if we might be better paying them to stop?
I long for the day when we see VMAX managed without worrying about what the internal engines are doing; when we set various performance parameters and let the array sort it out. When we pay for performance and capacity without worrying how the system gets to it.
There is at least one amusing part of advice in the article tho’ and it although it is well-argued and there appears to be good reason to do so; you still should keep the FC-tier on RAID-1 mirrored disks…Nothing really changes in the world of Symm!
The irony actually is that both with VNX and VMAX has either gradually been removing the “nerd-knobs” for the last decade or so (example = on VNX2 you’re not allowed to manually set read/write cache %’s even if you want to) or pushing towards best practices which recommend leaving things more automated than manual.
Doesn’t Isilon have a lot less nerd-knobs?
And I think its by design – make it more “appliance-like” and just add nodes to scale the storage capacity.
Nerb-knob fetting correlates to the dynamic capabilities of the technology. Disk lacks IO responsiveness for today’s data sets, thus lot’s of tunables for the ‘flash performance accessories’ that surround slow disk.
The market is realizing disk for applications is dead. it’s time for the technologists to stop trying to mask the problem and move to a better storage medium.
Martin, thanks for the pingback. At the extreme ends of the spectrum, I think there are two different kinds of VMAX customer — the oldschool storage admin, who wants access to every internal knob there is; and the infrastructure generalist, who wants to be able to provision & manage storage by service level. As we continue to improve VMAX usability, I think one of our challenges is how we cater to both of these types of customers (and those in the middle), without alienating one or the other. Our current vector of layering SDS / control-plane virtualization on top of hybrid arrays is, IMHO, the right one — it allows customers to either manage the knobs themselves, or use an SDS layer to manage the knobs for them.
Vaughn, I think there may come a day when disk for applications is dead… but with the cost of flash today & with budgets what they are, we need to offset the cost of flash in some way. For all flash arrays like Pure, XtremIO, etc — we offset the cost of flash with things like cores & dedupe software. This works great for workloads that dedupe well, or for small datasets with high performance requirements — you can fit heavy workloads into a small amount of flash capacity, and there isn’t much tuning required. Large transactional datasets that don’t dedupe well, or workloads with high reliability/availability/serviceability requirements, are still most likely going to be better off on a hybrid array like VMAX.
I’m going to speak generically so as to no inflame a vendor’s passions. I worked at a client site with several PB of provisioned storage and speaking to the main SAN admin, he does no tuning. He has no time to tune, the staff is too small. So their storage back-ends are large pools and auto-tiered. When they do a refresh , they won’t be interested in how to tune to get the most out of the storage frame. The knobs are something they are very much not interested in. Now maybe they are unique – but think about how many PB per single SAN admin today. No one will have time to tune. Check out the number of ads for storage admins – the positions are going unfilled. I’ve been contacted several times in the last six months about being a storage admin with “bennies” – and it isn’t my main skill set. Personally, as attractive as it might be to architect and manage storage, I shy away as the workload and dangers are very real. At another client site, the SAN admin replicated in the wrong direction – wiped out many VMs. Bye-bye SAN admin. He was probably tired, all the weekend work … ya know.