Latency
I'm with people who think NVMe needs to be as close to the CPU as possible to get best benefit. I'm willing to listen, but will be surprised if any solution sticking ethernet in the stack actually makes best use of the technology.
48 publicly visible posts • joined 17 Feb 2014
^^ They're hiring like crazy at the moment. I think you're deliberately skewing flexible working opportunities to create a rather twisted picture which suits your own narrative. If there were lay-offs in the company, I suspect El Reg would be first to report it and we wouldn't find out via some snide 2 bit commentary written in the margins by an anonymous account.
I lost a stone the other month ... where is my dedicated column in El Reg?
Interesting observation regarding NetApp.
When they don't change - they get slated for not changing. When they do, we get commentary that they're desperate. I'm not their biggest fan but I think I'd like to see a little less slap happy commentary about them.
This was always going to happen and they would have road-mapped for this from inception.
It's barking to suggest that a modern storage company would have recognised the value of flash - but only built their core proposition around it working with yesterday's technology.
Nimble will do hybrid better than most whilst spindles are viable and then they will deliver all flash at least as good as the competition in parallel.
They're obviously undertaking an infrastructure refresh and delivering that back in to the business as a managed private cloud service. It provides a different consumption model for the various elements of the business to adopt.
In the 90s and 00s - IT departments were tin buyers. Now they're turning in to service providers and I guess NATS are the same.
If VMware were a black hole .. here are the storage monkeys drifting silently towards the singularity of eternal lock in:
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Cras leo nisi, vehicula a tincidunt non, interdum a lectus. Nullam felis nulla, dapibus commodo arcu at, vehicula malesuada eros.
RE: "Let's reinvent the wheel!"
I thought that ZFS on Linux ran in user space - hence not fully integrated with the kernel?
And if bcachefs does things better, why not? BBC do TV - should Sky not do TV? BT do 'internets' should Virgin not? What other things are being undertaken by multiple people that are worthy of ire and scorn? Perhaps we should only have one Linux distribution?
I suspect that bcachefs will offer new opportunities for granular levels of caching. The fact that it's incorporating things such as snapshots is purely because these are the new baseline for file system requirements.
I can see the sense behind this approach. As new bright young things come on to the market, there will be a general assumption that somehow they're automatically fused around a core of innovation that makes them 'better'.
Any established company needs to batter these people back in to their box to keep a sense of perspective. It's their duty, hence Matt has to put his boxing gloves on and start swinging.
It's been a real struggle watching NetApp over the last few years, especially watching CDOT crawl out of the mud. Many of the problems it solves can now be solved elsewhere, so it devalues the impetus to move to it.
However .....
Every time I look at the storage market, I find that every vendor has multiple weaknesses but few have the breadth that NetApp have. Certainly none of the new guys. The potential to go single vendor end to end removes a lot of headaches.
RE comment above "Data is a 900lb gorrilla and she don't swing from silo-to-silo with any great ease" .. I can migrate my data one VM at a time through the hyper visor across multiple storage silos. As we used to say when we were kids 'whatyoutalkinbout Willis?'
I'm curious here .. do modern day storage tests accurately reflect what end users find valuable in products such as HCI? (genuine question - not cynical rhetoric).
Most storage vendors these days seem to be able to shoehorn in enough flash to crank up IOPS and stamp down latency to the point where I can reflect on the bigger picture.
It's the the scope of the product - the flexibility - the breadth of relevant features (not bells and whistles but the things that deliver real value). So when Nutanix can deliver storage efficiency features like compression / dedupe and then come in with cluster wide erasure coding etc, that's money back in my pocket. And it's hard to properly metricise things like the ability to run seamless non disruptive code upgrades - or management simplicity or speed of deployment etc etc. Same goes for EVO:RAIL ..
If these features are brought in under the testing microscope then great - but if it's all about how many sequential 64k blocks I can dump out to disk then the test warps the total value of the product. So there are legitimate reasons for getting the measure of the tests right - otherwise they get reduced to corporate propaganda paper waving exercises.
Vendor hair pulling ground fights are so far away from the real world end user conversations that I'm used to these days. This whole debate does have an air of the early to mid 2000s about it to me.
"Nutanix aren't offering much that, say, Simplivity don't" ..... As this filthy thread is already littered with profanity - I'll just add the word "Bollocks" here in response to this snippet.
Other comments that fall under this classification:
"It needs 3-4 years more before it's truly ready" .... says who and why? Justify your fantasy timescales here. 3-4 years in IT is a lifetime these days.
"but not one of Nutanix's customers has gone out and binned the ESXi hosts and replaced the netapp filers they already had" .... And you know this how? Are you some kind of infrastructure spy / ninja? ... Companies are buying new HCI and leaving their old crap in the DC because, what? They have money to burn on escalating maintenance of old kit?
Yes - Nutanix have to step up to the plate and join in with the whole storage testing malarkey in order to join in with the obligatory public chopper measurement (and they should get on with this sooner rather than later) but suggesting that they could still be irrelevant is frankly ridiculous.
Wow .. this sounds too good to be true .. and they're using what for the storage element? Say again? I thought you just said Storevirtual? Haha.
No .. really, what are they using?
Oh .. it *is* Storevirtual. Does it have any new features? No? Oh .. Okay then.
Why the hell do we outsource our analytic thinking to other people. Is time so limited that I'm not prepared to make a judgement call unless it's underwritten by 'Gartner'.
Create your own magic quadrant and place yourself in the 'visionary' section. Stick two fingers up to chart wizards.
Can you imagine NetApp going HC .. after spending a decade cooking up Clustered Ontap (or whatever it's called these days)? Affordable, flash enabled x86 web scale would undermine everything that they hope to achieve with CDOT. They're still in the zone which says that they can win with what they have.
They could have made a proper play with software defined storage and they've had the assets for some time (Ontap Edge anyone) but they've simply not wanted to go there, probably because of the potential impact to their traditional business. And if I were Cisco - I'd be looking to the future rather than throwing my cash at a company that appear to be in reverse - or at least wait to see if they can turn things around.
Over simplified? Yes.
Further from the truth than "People won't budge because they don't want to lose bells and whistles" ?? Not really.
This statement assumes that companies and their practices have evolved with the product rather than away from it. What do you think? Are people still intent on doing things at the infrastructure tier or are they perhaps solving problems in the hypervisor or in software these days? What matters if it's the latter? Bells and whistles or price & performance?
Like it or not, people will evaluate their options if they face a change. The 7 to CDOT transition is not comparable to the Trad to Flexvol switch. Why? Because the market and the quality of the competition 10 years ago were entirely different. NetApp were still fresh and WAFL felt unique. Now they're fighting a war on multiple fronts and customers have a wide range of credible options to chose from.
Georgens: "It is unlikely that customers will adopt competitive technologies that have fewer features and require even more complex migrations when compared with clustered ONTAP."
If the alternative does the bare minimum, is cheaper and flies like shit off a stick - trust me, they will adopt the competitor.
Now we're cooking and SDS is looking hot!
Personally, if the economics work out for this model and the features are there - then we only have to deal with the FUD canons that are aimed at the exec level before the war against the monolithic status-quo peddlers is won.
After I've spent a day or so rolling around naked in all the dollar bills I've saved, I'll dance on their graves at 1,000,000 ops a second.
"Broad portfolio of choices" .. ughhh I'm already sick of hearing this. How about "large range of plugs to fill the holes"? It's more gritty and down to earth.
The reason I find it irksome is that a couple of years ago the brag was "one OS fits all" and we were all laughing at the mess of EMC products needed to do what Ontap could do. Now it's a strength and not a weakness.
Let's break this down.
Flash is hugely disruptive and important, delivering extraordinary levels of performance. NetApp admit that CDoT is crap for flash. Their current strategy is to get people to invest in CDoT (or else the last decade of RnD is a complete waste) and operate in a non flash friendly environment for the next 4,5 or 6 years? And what - pretend that flash isn't happening?
You simply can't charge their prices and be all about the capacity .... because the moment you spend the big bucks and the thing doesn't perform, there are some awkward questions coming from the other side of the finance directors desk (read awkward questions as hairdryer moments).
This feels like a car crash in slow motion to me.
I hope you all feel that I adequately broke that down, as per my heartfelt promise at the start.
Chris is spot on with the extended analysis here. I wish things were different for the big N, but they can't turn back time and change strategy now. Perhaps if they had taken a snapshot before buying Spinnaker, they could have rolled back and started all over again. Or possibly cloned from it and investigated multiple strategic outcomes (in a time and space efficient manner of course).
Okay. I think you’re missing my point a little. For the record, decade + user of NetApp and long time fan. I’ve never recommended another top tier vendor in my life … but
You -[ “It doesn't cost any more to run cluster mode than 7 -mode aside form the cost of the cluster interconnect switch which is only needed when you have more than two nodes which many customers do not.” ]
I’m not saying that the CMODE or CDOT or Cluster Mode has a higher book price. My point is this, WAFL is getting on a bit now. A decade+ is a long time in IT. When you take a system like that and continually adapt to changing hardware and use case scenarios, you’re unlikely to match performance of systems that are designed specifically for the latest thing. Case in point, SSD. Are you telling me that Ontap/WAFL is the best solution for a hybrid disk array? Raid striping transient data in the flash pool / dedicating disks to parity / burning CPU cycles calculating parity etc? No, not really a great thing.
Is CDOT a good proposition? I like it in some respects and there are good features. But things like QOS for storage are not unique to this vendor. Take a look at Solidfire for a system that does it even better (mins and max – not just cap). NetApp cannot rely on bells and whistles to be winning key differentiators.
My main point is this – (and this isn’t just a dig at the big N because I’ve been a fan for years) – is WAFL able to scale with the hardware and deliver significant performance improvements over the next, say, 5 years? Looking at the FAS8240 series benchmark results on this site, I would say not. 30,000 iops more than a FAS3170 (and 30,000 IOPs in this day and age isn’t a huge deal for some of the new vendors coming online). In 5 years – that’s the performance improvement? The flagship storage node, beaten by a 5 year old 3Par array? So why not have doubts about what WAFL can deliver here. And if you follow that thought through - and consider binding yourself in to CDOT for the next 4 years – it then becomes a comparatively expensive proposition. All this at a time when the market is abstracting the bells and whistles away from the controller and sticking them in to software which can only lessen the demand for those features on array.
If you feel different about it , are happy with the performance and features then that’s great. You’re the master of your wallet – spend it how you like. Personally I’m not convinced and my gaze is drifting to the Nimbles and Tegiles that are materialising, and even more basic, cost effective server storage systems underpinning hypervisors and Openstack or serving software based storage. These are the game changers. Not CDOT although I wish NetApp best of luck returning to form.
It's a rocky period for NetApp. Ontap 7 mode was a great story but I'm not so convinced about the cluster mode situation. it's no simple transition and the benefits are questionable. Plus, to enjoy the benefits of a cluster mode world is going to cost you.
I feel that they've been massively wrong footed by industry trends. They've been so focussed on delivering cluster mode that they've missed the newcomers emerging with systems that are taylor made to work with flash (which WAFL ain't). Perhaps that's why MARS/Flashray feels 3 years late. Perhaps the cliff edge prospect of transitioning from 7 mode to cluster mode has given people time to consider their next step. If I were any other vendor, I'd be smelling blood right now.
Unless there is a fundamental change in their offering. I can't see where they will comfortably fit in the future of storage. They'll be squeezed at the low end by x86 scale out / abstracted high density storage and drained of life at the top end by all flash and hybrid vendors - many of whom will deliver IO without demanding extra money for licensed features.
As storage intelligence moves in to the hypervisor and VSAs - we need a simpler offering at the back end. One that is predictable & cheap. Can NetApp fit that bill?
Scale out clustering .... Because the only way to get any semblance of performance from WAFL is to source your volumes from separate aggregates on multiple HA pairs and stitch them in to a single name space. Nice, affordable solution that! I must buy me one, right after I strike gold.
I don't agree that the 'no tiering' strategy caused their current issues. The main problem, imho, is CDOT.
They've spent so long trying to bring this to market and make it 'a thing' that the industry has moved on. Now they're left to monetise years of development and convince their customers that they need to switch. I'm guessing that's not been a success.
The NetApp E-Series uses a method called DDP (Dynamic Disk Pools) to rebuild from multiple drives in order to greatly reduce rebuild times (x 8) so this isn't so much of an innovation. However, if GPFS can do it much cheaper, then it becomes interesting.
However, there are also plenty of other solutions that detect failing drives and pre-copy the data to spares so when the disk does go, it's ready to stand the next disk up in an instant. That should take the fear out of large capacity parity rebuild times.
Lots of noise comparing the new top tier to the 5 year old 3Par system.
What concerns me more here is that you can get 70% of the comparative IOPS on a FAS3170. Ancient mid range controller / top tier modern controller. I would have thought that if you stepped up to the next tier (two if you count the 6000 series) and added 6 years of hardware development that you'd get a more significant improvement.
If the license costs are still tied to the tiering, this makes it even less appealing.