Let... the SAN shine: 2013 – the year of virtual storage area networks
A look at the players who will fight the storage battles of 2014
Sysadmin blog Converged infrastructure – the trend of using one piece of kit for some pretty varied tasks, including networking, computing and storage, software and automation – really took off in 2013, mostly thanks to a group of robust virtual SAN players entering the market.
Barring an unexpected new entrant, 2014's big names in the vSAN space look set to be four: VCE, Nutanix, SimpliVity, and VMware, though I believe Maxta's willingness to address the mass market will allow them to catch up rapidly.
VCE is the original converged infrastructure company. It is the direct precursor to modern technologies and comes with a completely different corporate ethos than that which has made the younger upstarts so popular.
VCE is owned by Cisco, EMC, VMware and Intel and can be viewed as the functional embodiment of conservative data centre design. Nothing VCE offers comes in "small" or "cheap". Even VCE's "entry level" designs require tossing several virgins into nearby volcanoes and "commodity anything" will not be in evidence.
VCE has the most experience in converged infrastructure but it goes about it completely differently than the other players. The smallest widget it ships is a 24U rack full of gear. It relies on traditional SANs, switching, Cisco blade servers and so forth.
The rest of the market has evolved away from this, moving towards commodity servers filled with storage and using virtual SANs that create a GFS-like object store across the various nodes to lash it all together. Compared to VCE's conservative approach, modern converged infrastructure drives costs down dramatically; it doesn't sacrifice resiliency but can't match VCE for raw throughput or low latency.
A VCE vBlock: impressive, even in picture form.
VCE is pure enterprise infrastructure for companies with deep pockets. Its sales staff, social media evangelists, and marketing folks are of the no-holds-barred "death to commodity" variety that price-sensitive SMB sysadmins (like myself) all too frequently get the urge to deck. It is its own world, separate from the rest, only rarely spoken of in the same context.
Nutanix's benefit seems to largely stem from the massive number of people it has hired. It has good technology, yes, and also seems to have hired up many of the influential types in the VMware ecosystem. It has 4 VCDXs - a number I know is set to grow - many VCPs, vExperts and prominent community members.
It is well capitalised with a dominant mind share and capable sales people. If you go to any event even tangentially related to storage or virtualisation, there's Nutanix. It has a global reach of sharp minds that are charismatic, energetic and (most critically) know their tech cold.
Put it all together and Nutanix gives the warm fuzzies. People look at it and talk about the technology, not about whether they are going to be there in a year or not. Nutanix are clearly playing for keeps and seem to earnestly believe that converged infrastructure will be the new black: it isn't interested in a $1.5bn IPO because it doesn't see itself as a mere component of the data centre. It sees itself as the data centre: just add some switches to provide physical interconnection and Nutanix will take care of the rest of your infrastructure all on its own.
SimpliVity is different. To hear its people talk about its product, it doesn't see itself as the be-all and end-all of the data centre. It is far more focused on making the storage side things simple, efficient, and - most of all - robust.
SimpliVity's play is less about great big, monolithic Google-style data centres with racks upon racks of nodes all lashed together into some converged meta-hivemind. Oh, it says it has the tech to play in that arena, if that's what you want, but it seems to be betting that cloud service providers and mega-enterprises are only going to be part of the equation in the future.
SimpliVity is all about putting a SimpliVity cluster here, and one over there, a dozen or so in the head office and then making storage, backups, snapshots and so forth flow between them with no fuss, no muss, and no effort.
Like Nutanix, it has great people. Unlike Nutanix, the real money only started flowing very recently. It has done yeoman's work in the shadow of the giant so far, but the latest round is really what's allowing it to go out and start snapping up charisma and certifications. SimpliVity is slowly turning into the omni-presence that Nutanix has become, but it sings a different tune with a different pitch to a different audience.
VMware's vSAN is the big unknown. vSAN isn't out yet, and nobody knows how it will be priced. From the little I've played with the beta so far, it seemed brittle, awkward and unfriendly. vSAN feels like a very VMware product. It is something you deploy to a given cluster. You enable, toggle tweak and then operate within very specific guidelines and supported scenarios. It is designed to supplement the VMware ecosystem, but very carefully balanced not to dethrone EMC.
Some of that could be chalked up to it being a beta. Some of it, I'm less sure. VMware employees discussing the product have seemed taken aback by the idea that someone might want to use it in a manner that doesn't conform rigidly to the VMware Hardware Compatibility List and whitepaper configuration guidelines. I'd go so far as to call some interactions attempting to address requirements they viewed as "edge cases" to be outright hostile.
Attempting to call the end product from my limited exposure is admittedly something of a crapshoot. What I've seen so far says to me that VSAN looks to have all the rigidity of a Nutanix or SimpliVity: a narrow cone of supported units and no love if you colour outside the lines. It doesn't look like it's going to quite have SimpliVity's data-obsessed focus, Nutnaix's obsession with scale or Maxta's adaptability.
Despite all of this, vSAN will be sold by VMware, and that name - and established support apparatus - means a lot to enterprises. By sheer virtue of it's name, VMware are also the only company that has a built in solution to the brownfield conversion issue, but more on that later.
Maxta is "the new guy": so new in fact that it was only last month that The Register got the exclusive launch review of its product. It has technology very similar to both Nutanix and SimpliVity, but the play is different again.
Nutnaix and SimpliVity both want to sell you their software wrapped up to the hardware. They want as much margin as possible - naturally, who doesn't? - and are doing so by driving down support costs through viciously restricting the hardware they play on.
Maxta will work on anything with more than four cores (and that's a pretty arbitrary restriction, one that I think doesn't need to be there.) I've tested it on all sorts of crazy configurations, and so far it does "just work". Despite my continued efforts, I have yet to succeed in breaking it.
Maxta pitches simplicity of install, maintenance, and "invisibleness" of the product in everyday use. All of this I can attest to ... but you also lose a lot of ability to tweak the gubbins underneath ... something that whitebox types enjoy doing all too much. Nutanix and SimpliVity don't let you tweak anything either - they provide you a wrapped bundle of hardware and software - but the target audience of Maxta is far more likely to do things the software currently doesn't handle well: adding disks, upgrading disks and so forth.
Still, it's early days for Maxta. I'm periodically testing its software and firing up feature requests and bug fixes. Unlike SimpliVity or Nutnaix, Maxta is something I can afford to have in the lab... its competitors have a minimum buy-in price that neither myself nor any of my customers can possibly afford.
This here is where Maxta's potential comes in: unlike the competition, Maxta acknowledges the existence of the "other" 80 per cent of businesses. They also don't have a one-size-fits-all approach. If I have workloads that are super-storage-heavy, I can just scale a cluster to have lots of spinning rust and a few SSDs for acceleration. No big: I get to choose the mix.
That means I am not buying a bunch of CPU/RAM/Flash/Whatever capacity that will go to waste. There are a limited number of SKUs to worry about, and the minimum buy in could be as low as the Maxta software itself: existing servers will probably work just fine, making brownfield deployments (especially in the SMB space) a lot more likely.
Being a software-only solution, VMware theoretically shares this flexibility. The two vendors will end up doing the dance over ease of use, price and the breadth of platforms they choose to support. In at least one of these areas – ease of use – Maxta would seem to have the clear advantage.
The Other Guys
The traditional players like EMC and NetApp have been buying up similar solutions in an attempt to compete. Unfortunately, both EMC and NetApp have been utilising these solutions as an attempt at margin support, rather than a true means to move forward into the future of next-generation storage. Until they stop defending their traditional SAN market so desperately, they won't be anything but bit players in the virtual SAN market.
There are other companies that exist in this space which could well surprise us all, or they could stay in their niche doing just exactly what they do an doing it well. Stormagic is one such. Stormagic offers SvSAN, pitched as a virtual storage appliance rather than the new buzzword-friendly "virtual SAN." They nonetheless have design wins and a small but loyal fanbase. They support both VMware and Hyper-V but don't control a lot of mindshare.
Stormagic's current offering is two-node solution that does active-active replication. Think Starwind's iSCSI SAN, but as a virtual appliance rather than an application to be installed on a metal copy of Windows. It is a fantastic branch office or SMB solution, but at present isn't a converged infrastructure solution capable of replacing an enterprise's entire infrastructure.
Many of these alternative players offer virtual storage arrays that don't quite fall into the "up to 32-node SAN" category as the others. Here you'll find things like VMware's own VCSA or HP's StoreVirtual (nee Lefthand) SAN. In a sense, these are the evolutionary precursors to the virtual SAN players trying to win the converged infrastructure space, they haven't gotten enough love to be able to scale size or IOPS enough to play ball.
Other software defined storage players exist as well, and can't be discounted. Nexenta is gaining a lot of mindshare, though they don't look to be heading towards a converged storage/compute platform just yet.
There's also Exablox to consider. Exablox has a simple-to-use object-based storage system that is eerily close to the other big players here. While they are focused at the moment on the "simple, scalable storage" market, it is a very minor R&D investment for them to take that same technology and enter the converged compute/storage market.
Exablox storage blox: even cooler looking in person
Many hybrid storage vendors (such as Tintri) also have the expertise to be players, if they should choose.
Scale Computing wants a slice of the pie, claiming VSAN and VSA approaches are dead in the water even before they've had a chance to get their sea legs, and I haven't even gotten to SwiftStack, the Hadoop-as-a-virtualisation-backend players, and $deity only knows how many other entrants.
IOPS is where this really gets interesting. Nutanix and SimpliVity have been at this longer than the others. While I haven't been able to play nearly enough with the units themselves to compare, reported numbers out of both companies wreck the competition.
Maxta's working on it and it honestly isn't far behind. The numbers I get out of the lab when using the Micron P420 PCI-E card border on ridiculous, but it handles commodity flash devices (like a pair of Intel 520 480GB SSDs per node) surprisingly well, offering far better performance than I would have expected.
Efforts to engage VMware about vSAN IOPS – especially the idea of IOPS as a differentiator – have provided mixed results at best. Wanting to get to 300,000 IOPS has been derisively called an edge case; if so, it's an edge case I'm seeing requirements for even way down here where the SMBs roam. If such requirements truly are an edge case, I suspect they won't be for long.
The startups also are more nimble than VMware. Deduplication, compression, WAN optimisation, snapshot management, backup streaming, and so forth are top-of-mind or even old hat for them. vSAN is a looming threat because of the cachet of VMware's name, but they take half a continent to execute a turn and are thus less likely to be capable of going toe-to-toe with any of the startups on features.
The selling point of any converged infrastructure play is that you can lash together a bunch of commodity hardware to keep the costs down. Nutanix and SimpliVity work with companies like Supermicro and Dell to provide the kit: they just slap their logo on the front and sell it to you with their in-house software and support.
All of the solutions discussed above achieve enterprise class robustness and redundancy. The traditional need for "enterprise-class" hardware is now the niche. The software on offer is robust enough to cope with the unreliability of cheap gear: this is the Facebook and Google model brought to the masses. (Quite literally, in fact: some of the folks behind these solutions helped develop precursor technologies such as Google's GFS.)
The loss of one disk - or even an entire node - isn't an issue. Go ahead, shoot any random node; the whole thing will keep on ticking. The various players differ on how many nodes or disks the system can suffer before there is data loss, but Nutanix, SimpliVity and Maxta have all proven remarkably robust. There is no reason to believe VMware's VSAN won't be similarly reliable.
Supermicro FatTwin: storage/compute balance of the future?
2013 saw storage get shaken up, down, sideways and diagonally. Hybrid upstarts like Tintri started clawing market share away from EMC, all-flash vendors like NimbusData edged ever closer to affordable, the flash caching space exploded and converged infrastructure received a healthy dose of virtual SAN.
2014 is going to see a lot of jockeying for position as enterprises use the availability of alternatives to explore new relationships or at least drive down the margins of traditional vendors like NetApp. This is not good news if you're EMC, but the challengers still have to convince enterprises to actually switch, and getting new designs into brownfield environments is hard ... but that, dear readers, will a topic that will have to wait until the next article. ®
Sponsored: Hyper-scale data management