As I pointed out in the DSSD post from earlier today – we’ve been looking at the disruptive effect that the magnetic media -> NAND transition would have on the business, and making big moves in multiple dimensions.
- With XtremIO, we entered the all-flash array market with a new disruptive architecture, and rapidly accelerated to the #1 position in the all-flash market, with incredible scale-out, simple operations, and always-on always awesome inline dedupe, compression and metadata services.
- With DSSD, we’ve brought a radical new disruptive architecture, and created a new category – rack-scale flash. A strange cool new super-charger for all sorts of workloads.
- With VxRail, we’ve brought to market a hyper-converged infrastructure appliance who is optimized for all-flash configurations.
Long and short – if it not sufficiently clear, we think that 2016 is the year of the all-flash datacenter*
*the caveat is that we (unlike others) acknowledge that we’re taking about transactional workloads. Capacity-optimized workloads still have some time before NAND crosses over for $/GB economics – even after compression/dedupe and other forms of data reduction. For perspective, Elastic Cloud Storage appliances are below $0.30/GB acquisition cost already at scale, and most Isilon configurations are around $0.50/GB – will be a while before NAND gets there, but it likely will eventually.
Now – XtremIO, DSSD, VxRail = not enough. The reality is that there’s a lot of pressure from customers and the marketplace to accelerate the “all flash” revolution for the architectures they depend on, with the data services they need for traditional workloads. They need it for Enterprise Storage, and they also want it for Unified transactional NAS and
In that same late 2013 strategic planning session that led to DSSD, led to XtremIO acceleration, led to a focus on the hyper-converged SDS market also had another big conclusion – highlighted with a big red box:
You see – haters are gonna hate, and people will claim all sorts of things – but this is execution of what we think is key strategically to win in the transition that is well underway…. And it’s not a coincidence that DSSD, VMAX All Flash appear on the same day. They are two parts of a multi-pronged strategy to lead the flash transition.
While VMAX All Flash can be debated whether it’s evolution or revolution – one thing that cannot be debated is that it’s a big deal.
VMAX All Flash is:
- A radically simplified Enterprise Array package.
- Radically invested in dense NAND, and leaning in on all-flash, always.
- The first enterprise all-flash array to be fully unified – a completely embedded enterprise NAS stack – not an additional appliance with a separate management paradigm. Great to see EMC doing this before many of the startups who are racing to this point. I guess elephants can dance :-)
- Radically changing the economics – and disrupting the classic business model of EMC and the enterprise storage array market with #XpectMore based on what we’ve learnt in the appliance market and listening to customers.
I also saw someone (a good friend who recently moved to IBM’s BlueBox team) make a comment along the lines of whether this was a big deal, and whether it was an innovation.
IMO, YES. While there are emerging areas of IT where there is a ton of innovation around new workloads, cloud native applications, open source ecosystems (all things where we are investing heavily – just look at the last few weeks of Virtual Geek), people shouldn’t kid themselves – and BTW, I think that’s where he was coming from (and I think he’s wrong).
I meet a LOT of customers – of all shapes and sizes. They recognize that they need to build new apps, in new ways. But there is a LOT inside their enterprise that they CAN’T re-write (or won’t because it’s not the best bang for their buck). Savings, simplification for that world – it is what funds the innovation at the customer into the new cloud native world.
While not as radical as DSSD, or an all-flash VxRail/VxRack – understand something simple: VMAX All-Flash and XtremIO are able to radically reduce capital, operational cost, space power and cooling. That’s really, really important.
And Tyler, for what it’s worth – that’s what enables/funds cool new projects and innovations right beside them for new workloads that never want to see a LUN :-)
Put simply - there are workloads that depend on traditional enterprise data services (think SRDF, SnapVX). There are enterprise workloads that count on 99.9999% available infrastructure – because they aren’t architected for application-level resilience. There are customers who have consolidated workloads. There are customers who want integrated enterprise transactional NAS and block together. There are workloads who are not ideally suited for dedupe, and only compression – and therefore raw capacity flash cost/GB and density/RU is very important. If those comments resonate with you - then VMAX All-Flash is for you.
I don’t want to minimize “enterprise data services”. It’s not for everyone, and it’s a long list – but for many customers, they are really important. Want tens of thousands of volumes? Need Mainframe/iSeries support? Need broad OS/clustering support – not just the most popular open systems – but the esoteric ones? Want VAAI and VVOL support at scale? Want active/active? Want multi-site protection? Want deep app integrated recovery models? Want rich QoS functions? That’s a small part of the list.
Now you can have them with an-all Flash enterprise array that delivers up to 4PB of Flash, up to 4M IOps, up to 150GBps of bandwidth, with the densest NAND packaging in this class of device – and still the 99.9999% availability that VMAX is designed for
In VMAX All-Flash, there are also a ton of flash optimizations – above and beyond rapid constant qualification and integration of the densest, lowest cost NAND SSD (as of writing, 3.8TB SSDs) in very very dense packaging (the only thing more dense is DSSD – but VMAX All Flash is the densest of the all-flash traditional storage arrays in the world).
Examples above.
I’ll call out a couple – the “type II storage architecture” (see here) with a single shared memory space with a massive DRAM-based cache (up to 4TB) with materially changed and further refined cache algorithms focused on all-flash persistence means:
- A huge effect on writes – as fast as NAND is relative to magnetic media – it’s still slow relative to DRAM.
- Write folding means that we can coalesce writes before they land on NAND.
The other thing that we’ve learnt through competition in the marketplace, and also through our XtremIO experience – we must, MUST simplify our packaging and licensing – this matters a lot.
So… we did :-)
VMAX All Flash is packaged in a ridiculously simple way:
- V-Bricks. These are the “scale out” mechanism – each comes with to 53TB of Flash Capacity. Each one that you add adds CPU, Cache, ports/bandwidth in a simple linear way.
- Flash Packs. These are the “scale-up” mechanism – each comes with 13TB of Flash Capacity.
- Simplified software. Software is included in the appliance model that people are showing they prefer, and more software is included. For example, there’s a simple “F” model which includes: SnapVX, AppSync for app-integrated replicas, and in coming soon, non-disruptive mobility, and compression. The “FX” model adds all the SRDF variations, Storage Resource Manager, ViPR controller and more
Perhaps most importantly – I’ll say it again (because customers have told me – as recently as last week, Asurion team I’m thinking of you) – they are NOT licensed by capacity, but in a simple appliance-based model. That’s a ton of value.
I want to be clear about this – it’s a huge disruption into the classic economic model of Enterprise storage – which traditionally has been “buy the hardware, license the software (by capacity), pay maintenance and support, which will ramp over time, triggering a new capital purchase”.
NOT ANYMORE HERE. XpectMore is the program that started with XtremIO, and when you get the appliance, you get all the software, and there is a flat lifetime maintenance model. We listen, we learn, we adapt. Asurion team, I aluded to the answer when I was in Nashville last week, here you go – and thank you for being a customer and partner.
Lastly – I’m happy to say that we are going to be ready to go within days with orderability in converged infrastructure form – there will be Vblocks that leverage the new VMAX All Flash 450 and 850 models. We are working hard to accelerate all the major tech step-functions in our converged and hyper-converged portfolio – because why build when you can get on to the business of consumption and doing something of value?
We’re disrupting ourselves – and we deeply believe that in 2016, to not have an all-flash design for transactional workloads – well, if you don’t, it’s just not a modern datacenter!
Would love to hear your thoughts!
Great article as always Chad. I've been a Symmetrix/VMAX customer for over 15 years across multiple companies and industries and I'm excited about this announcement. Having an easy to consume all flash VMAX model is a BIG DEAL. I blogged a little about my take on the comments around the all flash VMAX from a customer perspective here: http://blog.mdschneider.me/leap-day-2016-vmax-all-flash/
Posted by: Matt Schneider | February 29, 2016 at 02:09 PM