2016 – the year 3D XPoint came down to earth from Planet Hype
And the year hyper-converged infrastructure put a dent in SANs
Storage Review in 2016 Storage in 2016 saw its on-premises SAN/filer array heartland assaulted by the public cloud on the one hand, and hyper-converged and software-defined storage on the other.
The now-classic dual-controller disk array went hybrid to keep storing primary data, but all-flash arrays are taking on that role and so the hybrids went all flash too. High-end monolithic arrays saw newcomer Infinidat make roaring progress as it took share with its low-priced and highly available InfiniBox array.
Secondary data was sucked off more and more to object storage, which itself saw a mass adoption of S3 as the preferred interface, leaving mainstream disk and hybrid arrays storing relatively less and less on-premises data.
The public cloud growth was unstoppable, with AWS out ahead, followed by Azure and Google, and then by Oracle and IBM – the only on-premises storage suppliers with public cloud service ambitions. The 451 Research group forecast public cloud storage spend to double in two years – with NetApp, HPE and IBM moving down the supplier rankings as Amazon's AWS and Microsoft's Azure clawed their way up.
Hyper-converged infrastructure was a storage hot point, with Nutanix having a successful IPO and powering ahead to lead the market. Cisco, Dell EMC, Lenovo, and HPE made hyper-converged products move as suitability for sales of servers inside a combined server, storage and networking sell exploded.
There were notable product introductions:
- Dell EMC's DSSD unit announced its NVME-accessed and NVME drive-using D5 array.
- Dell EMC re-engineered its VNX and VNXe arrays to produce the Unity line.
- Dell EMC introduced an all-flash Isilon scale-out filer.
- Cisco introduced its HYPERFLEX hyper-converged system, and it introduced a storage server version of its UCS line.
- Primary Data started shipping its DataSphere storage silo-converging product.
- SpectraLogic announced the biggest, baddest, largest tape library in the world – the TFinity ExaScale edition, with 1EB of capacity and support for LTO, IBM and Oracle formats.
- Amazon went into physical storage with its Snowball 80TB hard disk drive enclosure transfer to its public cloud from customer data centres, and its 100PB Snowmobile trucked shipping container of storage.
One notable non-event was Dell EMC's converged systems business deciding not to change vBlock servers from Cisco UCS products to Dell PowerEdge ones.
The most intriguing startup was Brian Ignomirello's Symbolic IO, which claims its technology runs database queries 60 times or more faster than other systems – even XPoint-using ones – through a different way of coding stored data and processing it. Micron flash storage exec Rob Peglar joined Symbolic IO as its CTO during 2016, giving it additional credibility.
Dell and EMC was the big event, a $67bn acquisition by privately owned Dell that saw storage industry legend Joe Tucci retire, and David Goulden's EMC Information Infrastructure business unit become the Dell EMC server, storage and networking organization. Michael Dell is now an entrepreneurial IT colossus alongside Bill Gates, Steve Jobs and Larry Ellison.
Load Dynamix and Virtual Instruments merged after VI ran into revenue growth problems. Load Dynamix investors provided $20m additional funding for the resulting Virtual Instruments company. The new company believes its ability to capture production workload profiles (via Virtual Instruments technology) and replay them in a test lab (Load Dynamix technology) will speed problem identification and resolution.
Virtual Instruments then bought hybrid cloud and virtualisation performance management technology business Xangati in November.
IBM bought object storage supplier CleverSafe and promptly put its software in the IBM public cloud, gaining first place in the object storage market.
All-conquering Nutanix bought VMware memory cacher PernixData at a price that we were told caused Pernix VC funders to take a hit in their investments.
Disk drive manufacturing boss WDC bought SanDisk and its assorted flash technologies and products.
Hyper-converged infrastructure appliance (HCIA) vendor GridStore bought DCHQ, renamed itself HyperGrid, and now delivers hyper-converged infrastructure-as-a-service (HCIaaS). The DCHQ tech, among other things, containerises existing traditional applications.
HCIA vendor Pivot 3 bought Nexgen, a flash array business with QoS technology, after it was spun out of SanDisk.
Cavium bought HBA and vendor QLogic in June and started making noises about NVMe over Fibre Channel.
Storage and Ethernet networking supplier Brocade bought Ruckus Wireless for $1.5bn in April as it looked for a new growth avenue. This walk up wireless avenue was itself overtaken when Broadcom bought Brocade for $5.9bn in November.
This signalled the end of the three standalone storage networking suppliers: Brocade, Emulex and QLogic. All three were acquired by larger, more broadly based businesses. The Fibre Channel SAN market was not big enough to support them.
All-flash array pioneer Violin Memory threw in the towel after a long and tortuous struggle with continuous loss-making quarters, a reverse stock split, NYSE ejection threats and then an ejection. It applied for Chapter 11 bankruptcy in December.
The misguided VC of the year award goes to Art Samberg, chairman and president of Pequod Capital, who bought 5.3 million Violin shares in June, before the NYSE expulsion, when they were trading at $0.62, implying a cost of $3.3m. They are now, in Violin’s Chapter 11 bankruptcy state, listed as worth $0.0462, implying a value of $245,000, a loss of around $3m. You win some, you lose some, eh Art?
Flash and storage supplier X‑IO – think ISE products – closed its Colorado Springs manufacturing operation and laid off employees in May. There is some hope that its technology, specifically its NVMe Axellio tech, might find a home somewhere. X‑IO had never become the success that financial backers Seagate and VC Oak Capital had wished for.
Crossroads sold its loss-making archiving products to StrongBox Data Solutions of Canada and is now focussing on patent licensing.
Quantum shut its Symform sync and share business down in June.
A non-death event was XtremIO, where rumors of its coming downgrade relative to all-flash VMAX inside EMC were strenuously denied, and the addition of file storage capabilities to XtremIO revealed later in the year.
At the CEO level:
- Hyper-V-focussed hyper-converged system vendor Gridstore ejected CEO George Symons in April. Chairman Nariman Teymourian eventually replaced him.
- Cloud storage gateway supplier Panzura changed its CEO from Randy Chou to Patrick Harr. Chou is now CEO and co-founder of cloud-based, security-as-a-service startup Nubeva.
- Mangstor CEO Trevor Smith was replaced by Craig Gilmore, who impressed the hell out of Mangstor's founder and board, despite having no storage technology experience. But then neither did ex‑CEO Trevor Smith, we were told.
- Qumulo CEO and co-founder Peter Godman transitioned to a CTO role as Bill Richter was recruited to be the new CEO and President.
- Triumphant Veeam co-founder and CEO Ratmir Timoshev stepped down, with William Largent taking over.
Ejections and divorce
EMC sold its content management business – think Documentum, etc – before its acquisition by Dell was consummated. Hopefully the new owner can turn this under-performing mongrel into a fresh puppy.
Symantec separated from Veritas by selling Veritas to the Carlyle Group, so confirming what virtually everybody knew: that it had failed to integrate the Vertitas storage products after buying it originally, with its security business, and there was no synergy between storage and Symantec's notion of security.
Imation's storage businesses were sold off by the owning Clinton Group, leaving an investing business behind. The Memorex trademark went in January with Nexsan's sale to private equity organised in December. With proactive ownership and a re-energised exec roster, Nexsan could grow in 2017.
HPE invested $10m in object storage vendor Scality in January.
Cisco invested in hyper-converged system software startup Springpath then introduced its Springpath-powered Hyperflex hyper-converged appliance and started a big push into the HCIA market.
Lenovo set up an alliance with Juniper to build converged, hyper-converged and hyper-scale data centre infrastructure products. It also partnered with numerous HCIA vendors: think Nutanix, Pivot3, SimpliVity, Atlantis and Maxta. It allied with Nexenta so its servers could run NexentaStor software, and partnered with Nimble Storage and its arrays.
Pure Storage and Cohesity agreed on a deal for Cohesity to provide converged secondary storage facilities behind Pure's primary storage flash arrays.
SanDisk allied with Red Hat so that red-hatted Ceph could run SanDisk's InfiniFlash flash-filled enclosures. So far InfiniFlash, with selling deals with Nexenta and Tegile, has not set the world on fire.
Where did the storage VC money go in 2016?
- Object storage and S3‑compliance evangelist Cloudian raised $41.3m in a D‑round.
- Security and file-sharing cloud storage gateway supplier CTERA raised a $25m E‑round.
- Datera raised $40m.
- Datrium and its server-powered storage ran a $55m C‑round.
- NVDIMMer Diablo Technologies raised $37m in a C‑round – and got itself a new CEO.
- DriveScale raised a $15m A‑round.
- End-point protector and file sharer Druva raised $51m in an E‑round.
- NVME-accessed array startup E8 had a $12m B‑round.
- All-flash array SW developer Elastifile drew in $35m in a B‑round along with a Cisco investment.
- Gridgain had a $15m B‑round.
- Hyper-converged appliance supplier Gridstore had a $19m C‑round.
- NVMe-accessed array startup Mangstor raised $5m in a C‑round; relative peanuts for a hardware/software startup's third round.
- Carbon nanotube memory startup Nantero ran a $21m F‑round; productisation taking a long time.
- Cloud storage gateway and data services supplier Nasuni raised $17.5m in an E‑round plus $7.5m venture debt financing.
- Flash endurance tech startup NVMdurance went through a $2.23m A‑round.
- NVMe-accessed array startup Pavilion Data Systems had a $15m A‑round.
- HCIA and flash array supplier Pivot3 raised $54.6m in a G‑round with bank funding.
- Scale-out filer startup Qumulo raised $32.5m in a C‑round.
- Secondary data silo-converging and data protecting Rubrik raised $61m in a C‑round.
- Hadoop and RDMS combiner Splice Machine raised $9m in its C‑round.
- Velostrata had a $17.5m B‑round.
- Software replicator Zerto raised $70 million in its E‑round.
That list represents a total of half a billion, $581.67m to be exact, in funding we know about that was pumped into startups and pre-IPO storage companies in 2016. That's not a great year with around $1.8bn being invested in 2013, over $3bn in 2014, and some $1.8bn in 2015. This is a significant reduction.
We figured a golden age of storage startups was coming to end in an article in September.
Nutanix ran its IPO in September and it has been judged a success. Its shares opened at $26.00 and are now trading at $26.86. In this IPO game that's a success, by the way. Pure Storage IPO'd in October 2016 with an opening price of $17.00 and shares now trade at $11.15; way to go Nutanix.
Magneto-resistive RAM startup Everspin filed an IPO in September. Think of MRAM as a kind of specialized XPoint-class non-volatile memory that doesn't have the potential application or marketing drive that XPoint possesses. Everspin IPO'd in October with shares starting at $9.10. They are now worth $7.84, having risen from a low of $6.33 in late October.
All-flash arrays became mainstream, with every incumbent vendor either bringing out new-design flash arrays (EMC DSSD, HDS A‑Series) or retrofitting SSDs or flash modules to existing arrays.
Nimble Storage completed its pivot to all-flash arrays with a successfully introduced product. NetApp's all-flash FAS proved popular. Pure Storage is the only all-flash array startup to make it to IPO so far and has remained a top-four all-flash array supplier. Indeed, it opened up its second major product line in March with FlashBlade, a rack-scale system using proprietary storage blades, for storing unstructured data. Pure's triumphal 2016 was marred only slightly by having to pay its way out of a Dell EMC lawsuit over IP.
Interestingly, Toshiba announced its FlashMatrix competitor to FlashBlade in August. This is an Atom CPU-powered realtime, scale-out, compute-plus-flash analytics engine. Toshiba has yet to announce how it will come to market, meaning channel, and there was no pricing or availability information at its launch.
NVMe-over-fabrics, with its server storage stack-bypassing RDMA technique, could be the last major technology development for on-premises array connectivity. Many startups focussed on the area; Apeiron, E8, Excelero, Mangstor and Pavilion Data Systems for example. Virtually all the incumbents said they would support the technology, including old-timers HPE and NetApp, and newcomers Kaminario, Tegile and Pure Storage. Dell EMC's DSSD unit introduced its D5 NVMe over fabrics-accessed array. The fibre channel HBA suppliers said they would support NVMe over Fibre Channel, providing an installed base upgrade path.
Server operating systems and applications need to transition to NVMeoF – a clumsy acronym – and the impetus behind that should strengthen in 2017.
SSD density-increasing 3D NAND became mainstream as well, with WD (SanDisk) and Toshiba on the one hand, and Intel and Micron on the other following Samsung's lead and shipping product. Intel announced its 3D NAND SSDs in March. SK Hynix also has shipping 48-layer 3D NAND product and is developing 72-layer technology. Toshiba and others started talking more about QLC (4bits/cell) flash technology as a possibly affordable solid state archive medium, where its low write endurance would not matter. Toshiba briefed attendees at the Flash Memory summit on a 100TB QLC SSD idea.
Even China has its own plans to enter the 3D NAND manufacturing area, having been rebuffed when trying to buy Micron and invest in Western Digital.
DataCore dominated the SPC-1 benchmark with its Parallel IO software and just blew the price-performance numbers away; sub-$50k DataCote x86 servers trouncing $1mn-plus mega storage arrays, and showing storage array IO waits had large CPU wait components as IO stack processing hit CPU bottlenecks. Oracle SVP Chuck Hollis said the SPC-1 benchmark and DataCore's tech was irrelevant to real IT life. DataCore's chairman said that was just sour grapes.
Latecomer Seagate finally entered the helium-filled drive era in January with a seven-platter 10TB disk. It said it would ship 12TB helium drives for testing later in the year, with 14TB ones in its roadmap.
Toshiba, the third disk drive supplier, did not introduce helium-filled drive technology, confirming it will be a capacity laggard. However, such drives might arrive in 2018, assuming Toshiba can get over the multibillion-dollar hit of its Westinghouse nuclear power plant projects in the USA, without re-planning disk drive investment schedules, or indeed selling its disk drive business.
Seagate says brand new heat-assisted magnetic recording (HAMR) drives will appear in 2017, with HAMR replacing perpendicular magnetic recording. Think 16TB initially? Infinidat sees 20TB disk drives on the horizon.
Just to show what it can do, Seagate filled a 3.5-inch drive enclosure with flash to produce a 60TB demo SSD. It's a startling idea but the cost would be prohibitive, any other SSD maker could easily do the same thing, and it basically seems like face-saving technology because Seagate is a bit-player in flash and needs to do much, much more. A partnership with SK Hynix has been rumoured but nothing has been said publicly by either company.
Our sense is that if Seagate chairman and CEO Steve Luczo doesn't get Seagate a proper flash strategy and chip supply partnership then toastdom threatens.
To add ignominy, Seagate's Kinetic disk drives, Ethernet-accessed drives with an object storage capability, sputtered without catching fire, while Igneous and OpenIO rejected Seagate's tech and introduced their own products, each plugging an ARM-based micro-server onto 3.5-inch disk drives. We'll see how they do this year.
Object storage made steady progress throughout the year, with HPE investing in Scality, Dell renewing a Scality OEM deal, and WD HGST unit introducing a new 14PB archive array. With Dell buying EMC then that organisation has in-house object storage technologies to measure against Scality's RING and 2017 might see a shakeout there.
As with its Nutanix OEM deal though, the level of customer demand could prove to be the deal's saviour – or the opposite.
IBM leapt to the top of IDC's object marketscape charts following its late 2015 acquisition of CleverSafe. The software is being used in IBM's public cloud and this, together with its FlashSystem are highlights of what has been the now-traditional decline in IBM's storage hardware revenues throughout 2016. However, IBM did put a new executive in charge of storage and he, Ed Walsh, might make visible changes in 2017.
Hyper-converged infrastructure was the dominant storage story in 2016. Nutanix's IPO was part of that. Dell EMC with VSAN, ScaleIO and VxRail/VxRack was the number-two vendor, and straining every one of Chad Sakac's sinews to overtake Nutanix. But it didn't get there and, indeed, Dell decided that the combined Dell EMC should carry on selling the OEM'd Nutanix XC system; customer demand was that important.
HPE launched its gen 2 hyper-converged product in March. It launched the even newer (gen 3?) HC 250 and HC 380 systems in December. SimpliVity added Hyper-V support in April. Even NetApp started talking about a hyper-converged system it was developing.
HPE is preparing for servers with enhanced and persistent memory capabilities, with Machine-developed technologies on the one hand and non-volatile flash and XPoint or ReRAM components on the other. It introduced a ProLiant server using power fail-protected Micron flash DIMMs in March, and the Machine project was splashed all over a customer event in November. HPE's Memristor technology seems as far away as it ever was, particularly with HPE Labs boss and Memristor evangelist Martin Fink retiring, and HPE signing a partnership with SanDisk over ReRAM.
Our sense is that flash DIMMs are still an early stage technology with no great server IEM wins by, for example, Diablo Technology. Our sense is that non-volatile DIMMs are going to be an XPoint-driven technology and not a flash-driven one.
Faster 32Gbps Fibre Channel, doubling 16Gbps FC speed, was introduced smoothly during the year but its impact was muted by the gathering enthusiasm for far faster array access using RDMA techniques and also by the rise and rise in virtual SAN-using hyper-converged systems.
The concept of Container storage flourished with, for example, NexentaEdge providing storage for stateless Docker containers. Portworx provided persistent storage for containers as did StorageOS and also Hedvig. But, by and large, storage for containers is still an early-stage technology with no clear winners.
3D XPoint had a poor year it has to be said, with continued denials by Intel that it was a variant of phase-change memory (PCM), and a drip feed of performance data that fell a long way short of the grandiose claims a year earlier that it was 1.000 times faster than flash, etc. Intel tried to escape this trap by saying it was talking about raw media speed but, having set the performance, density and endurance bars so high, its original marketing now looks to be hysterical and quite alarmingly inaccurate or imprecise.
Whatever the reasons XPoint NVMe drives now don't look so far in front of everything else as to render ReRAM or PCM drives pointless, and XPoint DIMMs, if using DRAM caches like flash DIMMs, won't be that far ahead of flash DIMM performance. We should see actual product this year and will get a better view then. So far, though, XPoint is marketing hype, and Intel and its partner Micron, have to show that their Optane and QuantX products actually merit the hype.
Parallel IO was promoted by DataCore and Data Domain. Bridgeworks pushed its TCP/IP networking parallelisation technology with radical transmission time shrinkage when TCP/IP is used for replication with, for example, NetApp SnapMirror.
Backup software supplier Veeam had an excellent year with revenues surging ahead. Both Barracuda and Commvault recovered from revenue dips by responding to customers wishes for public cloud-based protection and also for appliances. HPC storage supplier Panasas had a so-so year, with several exec departures early on. Teradata had problems.
Tegile chopped a few staff in February, and so too did Atlantis as well as Data Gravity; Paula Long's startup succumbing to financial gravity pressures. Late in the year NetApp laid off several hundred people as part of its efforts to grow again. Veritas also went through a pink slip exercise in December.
The on-premises SAN/filer heartland is now in the midst of an all-flash transition for primary data and vendor revenue growth looks to be primarily about taking share from other vendors rather than greenfield site sales. The ways of doing this include better analytics, faster flash and better hybrid (public-private) cloud support.
HCIA vendors, secondary storage silo convergers, storage SW-only suppliers, cloud storage gateway file sharers and protectors, and object storage suppliers all see the classic on-premises SAN/filer array as a big, fat, dumb, complex and complacent carcass they can collectively rip to shreds.
But there is hope for the SAN and filer suppliers. The key to its survival is dragon-slaying, with the two primary dragons being hyper-converged infrastructure and the public cloud.
With product enriched by all-flash and scale-out technology, by NVMe drive and fabric access, by better QoS and analytics, by object storage linking and public cloud back-end access, it can offer a better-than-public cloud storage experience overall and beat HCIA deployments as well. At scale, shared storage is a better bet than hyper-converged storage. As long as the HCIA vendors are trapped by that perception then modern SANs and filers have a chance.
But if the HCIA vendors show they can support large scale deployments and the public cloud suppliers continue surging then the on-premises SAN/filer outlook is bleak. See you this time next year to see how things turn out. ®