You're only young but you're going to die: Farewell, all-flash startups

Following news of its restructuring there will inevitably be layoffs at X-IO’s Colorado Springs facility. We hear that the company is working to get staff employment with other companies. The situation is fast-developing and X-IO’s concern is its employees and not the media, so we'll hear more later. Our interest here is not …

[Pascal Monett]

So, Flash is like search engines ?

There's Google, and there's the rest. And the rest need to invest billions in data centers if they want to have a chance. Those who can't become Google-scrapers.

Flash appears to be the same game. The incumbents have the investments made and money to invest more. Newcomers are learning that the entry ticket costs more than they hope to earn.

This is pure capitalism. No amount of lobbying or backroom deals is going to change this market. You gotta come with the billions, or you can leave while you still have a car to leave with.

[Anonymous Coward]

Re: So, Flash is like search engines ?

Actually the article is a bit misguided.

It starts with a logical fallacy, then, using that fallacy, proceeds to an argument based on flawed assumptions.

This is the fallacy:

X-IO, Violin and Kaminario were never really successful, or really differentiated. And Violin and Kaminario were around for a long time before the current crop of AFAs and still weren't successful.

Nimble and Tintri arguably have value add far beyond all other vendors in certain areas. For instance, the analytics in Nimble's InfoSight are truly impressive, or consider the serious per-VM capabilities Tintri has.

Like with anything, there are different customers for flash.

Some just want faster than spinning disk, and are happy with arrays that offer good features, ease of use, good performance and good economics via data reduction for general workloads (Pure and Nimble for example).

Then there's the class of customers that cares about nanosecond-class latency and does not care about performance or space efficiencies or, indeed, cost.

DSSD is good for them.

But saying anyone that doesn't do NVMe fabric is doomed is a bit shortsighted.

The vendors that do NVMe fabric are ultra niche at the moment, low on features, and in general not quite baked yet.

And thinking everyone else is ignoring NVMe fabrics is also misguided.

[Anonymous Coward]

Re: So, Flash is like search engines ?

I wish I could upvote this a million times.

[Anonymous Coward]

Re: So, Flash is like search engines ?

Yes, and this always happens with storage... the last batch of next gen storage arrays were Compellent, 3PAR, XIV, Pillar, etc. They were all acquired and the rest went away. The last time there was a break out array which became a stand alone company at scale was NetApp.

[Anonymous Coward]

Re: So, Flash is like search engines ?

I believe the article is spot on. Even though the X-IO's of the world have been around longer than the newest crop of AFA players, they were never able to break out and gain respectable market share. Investors kept throwing in dollars in hopes something would change and that their initial investments would not go down the drain.

Why is this? Because XIO, as well as today's AFA vendors, are now competing with a totally different technology shift: server-SANs. Before you tell me that there's been this technology around forever and a day, let's consider why this technology is hitting hard and fast just now: SSD. Period.

Next technology to shift is memory-based storage, but let's stick with SSD today. I'm speaking in terms just about anyone can understand...it's fairly simple to see the logic.....

SSD's are growing to capacities never heard of before. Like DRAM, the larger the models, the lower the cost of previous models. 16TB will soon be common....growth is SO much faster than spinning disk.

Customers that could not afford server-SAN before will be able buy it off the shelf for anything performance-related (short-term outlook). Long-term outlook is that ALL data will be in this tier.

If you could put all of your capacity and performance inside the compute within budget, why would you not??? THAT is the question and also the reason external array vendors will continue to feel pain until they're bought, adapt, or go under (XIO).

My two cents....

[Killing Time]

Yes I get it, they could be said to be on a Highway to Hell and can hear Hells Bells....

[Anonymous Coward]

Tintri

As a loyal and devoted Tintri customer, I have an objection to lumping them in with Tegile and Nimble. Tintri makes a storage device exclusively for virtual environments which is designed to deliver high performance and simplicity. That niche may not sound exciting, but after endlessly dicking around with block storage protocols, it's my experience that Tintri offers something that literally no other vendor can provide: speed and simplicity in one package. You can unbox a Tintri and have it racked, cabled, powered on, and serving data within half an hour, and then you basically never need to touch it again except to perform occasional OS updates, which are in-place and non-disruptive. Compare with EMC and NetApp, where you can spend months planning your deployment and then have to have specialized knowledge to keep the devices running and to operate them, driving up TCO immensely over the period of ownership.

Basically, Tintri is the only storage vendor that I've found whose product doesn't come with any gotchas. Not saying that there aren't others, but that's the only one I've dealt with.

[johnward]

Re: Tintri

As a Tintri customer, I don't see lumping Tintri in with other storage either. They only do one thing and do it so well. Tintri marketing asks what are all the cool things you do with Tintri. Their storage awareness and other features are nice and shiny, but the real money is that they just work. They work fast. I don't spend time working on them.

[oli_from_germany]

Re: Tintri

This is BS. Setting up a FAS takes 30 minutes max., do not have to touch it again. Disclaimer: Employee - but still true. I don't know who is spreading this FUD but I just replaced a Tintri at a customer due to high maintenance and lack of feature at the Tintri side.

[johnward]

Re: Tintri

Apparently "employees" are spreading FUD. I'm still waiting for professional services to setup our FAS based cDOT cluster. I am not even supposed to power it up until they get here. Professional services is allocating a full day for this, not 30 minutes. After it's setup the FAS will not know where each and every IO is coming from. It will not report which fraction of latency is the host, network, or storage with drill down to flash, disk, contention, and quality of service. The FAS's ability to eliminate noisy neighbor issues will still be zero.

[bitpushr]

Re: Tintri

Yeah, that's demonstratively not true. I'm sure the formatting will get screwed up, but:

dot83cm::> version

NetApp Release 8.3.2: Wed Feb 24 03:29:11 UTC 2016

dot83cm::> qos statistics latency show

Policy Group Latency Network Cluster Data Disk QoS NVRAM

-------------------- ---------- ---------- ---------- ---------- ---------- ---------- ----------

-total- 430.00us 87.00us 113.00us 83.00us 98.00us 0ms 49.00us

User-Best-Effort 430.00us 87.00us 113.00us 83.00us 98.00us 0ms 49.00us

That's my lab system, running cDOT 8.3.2. QoS is free and always on. Disclaimer: NetApp employee.

[Trevor_Pott]

Re: Tintri

HAHAHAHAHAHA Netapp.

[Justin Clift]

Meanwhile, on the super low-end... :)

Even FreeNAS (FreeBSD based NAS) has people playing with adding Infiniband support to it. Note, self promotion. :)

https://github.com/justinclift/freenas-infiniband

[Anonymous Coward]

Re: Meanwhile, on the super low-end... :)

Well hit! Precisely the OS and vendors I was looking at here.

[Justin Clift]

Re: Meanwhile, on the super low-end... :)

Cool. :)

The native ethernet support for Mellanox adapters seems to "just work" (after compiling in the driver). It's been pretty much trouble free.

IPoIB mode though... not so much. Kernel panic when switching from Ethernet → Infiniband mode (post boot - still chasing this bug down), and connected mode IPoIB exhibits weird behaviour too at the moment.

I think the Infiniband mode problems are rooted in something to do with the TrueOS base (not pure FreeBSD) that FreeNAS is built upon, as the same drivers on a FreeBSD base are pretty flawless. I'll be digging into this more over the next few days to see if I can figure out WTF is causing the issues, and then get it fixed. :)

If someone's just after Ethernet support (10GbE/40GbE/etc) out of Mellanox adapters though, it seems to be already good enough. Naturally, test the living heck out of it before deploying to production, just to be safe. :D

[Anonymous Coward]

No

I don't buy this. I work for an AFA vendor and I just don't see real customers asking for NVMe fabric-attached arrays. They have existing HDD platforms connected via FC or iSCSI - and that's exactly what they are looking for from their AFAs. I've never even been asked about it in meetings - and I've met enough customers to make that a reasonable view of the marketplace.

Don't confuse futures with what's required today.

[future ready]

all arrays will die

How many workloads won't run happily on an all-flash hyper-converged platform regardless of the brand? Seven and fifteen TB SSDs are launching at the end of this year, capacity and cost per GB for flash are no longer an issue. If you are a company that only produces storage arrays and you are not profitable now, you are dead. To think otherwise is foolish. I work for a major storage array vendor, but we also have other options that are server centric thankfully. I give all storage array start ups two years tops regardless of how well integrated they are with the hypervisor. The hypervisor is the storage now.

[random_graph]

The Better Mousetrap rarely wins

I tend to agree with Mellor.

The external storage market is optimizing along the guardrails:

One path: Arrays for transactional workloads, with the purchase led by $/IOPS and solution integration. This is NVME, AFA, and memory-based solutions.

The other path: Storage for persisting unstructured data cheaply and efficiently. This would be scale-out NAS and RESTful object.

And the middle - all the run-of-the-mill enterprise apps running in ESX - will be slowly captured by VSAN/Hyperconverged (including HDFS), especially as VmWare and MSFT duke it out for the full storage stack, blowing away the remaining mid-range array vendors in the process.

Yes Tintri and Pure make a great mousetrap, but in a maturing market, that can never be relied on as the ticket to survival. Customers not going to VSAN/HC will take the low-risk path of EMC, HP, IBM because their products are "good enough". Pure and Tintri, in an effort to show positive cash flow, will eventually have to cut back on cost-of-sales, and then getting the n+1 customer gets harder and harder.

[Anonymous Coward]

C'mon Reg, you know better than this...

There is a TON of empty space that is being left out in the statement of "EMC leads the all-flash array market from a revenue sense, IBM is second, HPE third and Pure fourth..." and I am quite certain everyone that reads this article objectively is aware of EXACTLY what I am saying!

For those that aren't aware, I will clear up the info that is missing from that statement:

1) All three of the "big 3" have experienced, at least, double digit losses for the last year in the overall storage business (whether that is hybrid, or all flash) to both the cloud, and competitors like (mostly) Nimble, Pure, Tegile, etc.

2) "...effectively closing off the market to..." is hilarious at best when you figure in point #1, and the fact that most of those competitors (Nimble, Tegile, Pure, Tintri, etc) are ALL experiencing growth even in the face of what the "big 3" continue to tell the industry.

3) When price, complexity, scale, density, and capabilities are figured into the equation, and even if you add ALL of the storage vendors into the same pool (either from an all flash, or hybrid perspective) the vendors that can TRULY compete and offer a solution that incorporates all of the good features, and eliminates the bad, becomes VERY small.

4) Hyper-Converged is still wet behind the ears. Folks are willing to incorporate other storage options into their infrastructure for specific workloads/use cases...but that is a FAR CRY from replacing their entire infrastructure and putting all of those eggs in one vendors basket. Especially when one has been around for quite a long time and not "taken over the world", and the rest have been here for effectively 10 min.

The IPO market being hard is undeniably true, and that will serve to pare off those offerings that have short life spans (at least as themselves), but also to prove out those that deserve to be where they are because of the products/technology that they offer.

As the title reads...The Register knows ALL of the above statements are true, as do most of the folks that play in the storage industry

Just my $.02

[Anonymous Coward]

Re: C'mon Reg, you know better than this...

Wet behind the ears? I am most certain you're not paying much attention the money being spent on HC....it's a LOT and still growing. The reason ANY vendor that sells only external arrays will die in the next 5-7 years is because of the size and price of SSD. Period. Sure there will be other technologies even more exciting than SSD, like XPoint for example, but as SSD's get bigger and bigger and the price/GB continues to drop, why oh why would you ever build an external array for any app that needs performance?? The only reason for an external array will be cheap and deep for 100's of TB or multi-PB capacity. Get your sales while you can...it won't last long.

[Anonymous Coward]

Re: C'mon Reg, you know better than this...

So we are to take you at your word that Hyper-Converged (HC) is wet behind the ears and leads to one vendor basket? You're kidding right? There are two players in today's HC market from a hypervisor standpoint: VMware and Microsoft (with next release, so yes MS is a little wet). Customers already run VMware....now you can put the storage WITH the HC/compute. There is NO basket to speak of...you're actually moving OUT of the basket which is the array vendor.

Go ahead and keep believing that HC is a fad, flash-in-the-pan technology. We'll see where the AFA vendors are in three years.

[elan]

panta rhei

well - wishful thinking

- the big guys are in muddy waters

--- emc/dell => many questions about productlines

--- netapp => revenue horribly declining

--- HP(E) => no margins, (free hw delivery bc of lousy data reduction values)

if you look at the market share numbers "others" is the growth value in it.

yes, market consolidation will happen --- but not now.

[Anonymous Coward]

The Elephant In The Room

So many of these articles about the current state of the storage industry seem to miss the point that the overwhelming market share trend is the growth in non-branded ODM storage to cloud players. This is causing a decline in revenue to traditional vendors. This trend is established and increasing. So surely it doesn't really matter whether you're flash or hybrid, large or small and have NVMe fabric attach or don't. If you're an enterprise storage player relying on traditional capex sales to end-users then you are in hot water because the market is moving away from you. Comparing features and who can set up an array the fastest seems a bit like fiddling while Rome burns. BTW - how old is Trevor Potts?