Big iron storage supplier Infinidat blags more o' that sweet VC cash

Moshe Yanai's Infinidat has gained $95m in third-round funding and wants us to know that it's not a debt-fuelled Silicon Valley extravaganza of a startup like others that have crashed and burned or gone through bought-at-a-discount acquisitions. Infinidat was founded in 2010, took in $80m around 2012, $150m in 2015 and has …

[NVMe or bust]

Infinidat vs AFA

No reason why AFA can not compete on capacity per tile floor with INFINIDAT - Flash SSD has already easily surpassed HDD capacity per volume. Cost/Price is a different story.

NVMeoF will nor change the capacity equation, but has the potential to lower response time.

[zbmwzm3]

Infinidat gets more money from venture capitalists..yep got it. Smells a bit like sponsored content.

[Lusty]

What's this all got to do with Big Iron?

[Anonymous Coward]

Tegile and Nimble aren't "really" AFAs: they were hybrids like Infinidat until recently

Infinidat is part of a larger wave of companies started in 2009-11 focused on hybrid block solutions: how to use flash as a cache with HDDs underlying it.

Tegile, Nimble and Tintri who led this wave have all pivoted to all-flash at some point or another, which ruined their entire hybrid marketing, which may explain (or not) why they all failed.

Infinidat is the last man standing that's keeping true to its hybrid roots.

Should Tegile/Nimble/Tintri have stuck to their hybrid origins rather than move to all-flash?

Discuss.

[Anonymous Coward]

Re: Tegile and Nimble aren't "really" AFAs: they were hybrids like Infinidat until recently

Sorry AC, you have been ill informed, or you have chosen to pass on FUD instead of doing your research:

- Tegile is a ZFS based array (with some secret sauce) that technically is media agnostic to a certain degree meaning that they need "spindles" but they are certainly NOT an "all flash array", nor did they abandon their hybrid platform in the slightest. Oh, and they didn't "fail", they were purchased by WD

- Tintri is a ground up design leveraging NFS and VMware (etc) orchestration on the storage side to simplify virtual machine deployment/management. Additionally, they are also NOT an "all flash array", nor did they abandon their hybrid platform in the slightest. While lot's of indicators point to the possibility of Tintri not being long for this world, they have not "failed"....yet

- Nimble Storage is a ground up design that moves all performance into the compute layer, and as such is completely storage agnostic as well. Their very first product was indeed (in 2009 by the way) an "all flash array", however, due to their media agnostic nature, they pivoted to a hybrid array which was a MUCH larger addressable market. While they do indeed offer an All Flash Array, they have certainly not abandoned their hybrid array either. Oh, and if being purchased for $1.2 billion dollars is "failing", I can guarantee there a lots of folks out there that would LOVE to "fail" that way!

Infinidat is a "me to" hybrid platform with large scale that wows potential customers with its 1PB marketing. Very capable storage, very large scale, but the performance is certainly NOT, nor can it ever be (unless RAM becomes MUCH less expensive) comparable to an All Flash Array.

Plain and simple, folks that can address ALL THREE requirements: Consistent Low Latency (Flash), Very Low Latency (Hybrid), and Extreme Scale (PB+) are the overall companies that will "win".

Right now Infinidat is addressing an available segment of the market that is getting less and less viable. As folks continue to tire of having silo's of storage (one for extreme capacity, one for low latency, etc) then they will have to make a decision that will be much more challenging to address in their current iteration...and putting 1PB of flash on a datacenter floor and not charging for it until the customer needs it is going to be VERY expensive!

Sometimes we need a magnifying glass to understand what real growth is: If I have 100 customers and next year I have 200 customers, that is a 100% growth pattern, and so on...but I am still irrelevant in the overall market until I have thousands and grow by thousands!

Should any of the above "failures" have stuck to their hybrids? The answer is "yes", and by the way...that is what ALL OF THEM has done! FUD is not, and has never been FACT!

Just my $.02

[Skives44]

Re: Tegile and Nimble aren't "really" AFAs: they were hybrids like Infinidat until recently

Hybrid arrays imply that the system is persistently storing data to multiple media types. INFINIDAT is not a hybrid array as it only persistently stores data to spinning drives. Writes and Reads come happen at the DRAM and SSD layer. Do your home work Einstein.

[Anonymous Coward]

Re: Tegile and Nimble aren't "really" AFAs: they were hybrids like Infinidat until recently

" Writes and Reads come happen at the DRAM and SSD layer. "

Writes I get, reads I don't. You can prefetch from spinning into DRAM/SSD but for random data access, the prefetch algorithms only get you so far.

[jamesrthrone]

Re: Tegile and Nimble aren't "really" AFAs: they were hybrids like Infinidat until recently

That's because you are vastly under-estimating the patented innovations the system leverages with regard to data placement and accessibility. Think "machine learning" algorithms rather than "pre-fetch". Those are absolutely critical in order to hit the >AFA performance numbers we're talking about here, or else, as you say, we would have reads coming off NL-SAS. That's what hybrids do - not us, we are a Cache Array. We use cache for host I/O's and NL-SAS for persistent storage. It is a modern, innovative design, and frankly one of a kind. Hope this helps.

[JCWCVG]

Re: Tegile and Nimble aren't "really" AFAs: they were hybrids like Infinidat until recently

Sorry AC, you have been ill informed, or you have chosen to pass on FUD instead of doing your research:

- Tegile is a ZFS based array (with some secret sauce) that technically is media agnostic to a certain degree meaning that they need "spindles" but they are certainly NOT an "all flash array", nor did they abandon their hybrid platform in the slightest. Oh, and they didn't "fail", they were purchased by WD

*It is rumored, that Tegile was purchased for as little as $55M. Here is what is not a rumor, employee options were worthless. Just ask any of them. Please do not confuse a sale as a successful outcome. See Nimble below. #Fail*

- Tintri is a ground up design leveraging NFS and VMware (etc) orchestration on the storage side to simplify virtual machine deployment/management. Additionally, they are also NOT an "all flash array", nor did they abandon their hybrid platform in the slightest. While lot's of indicators point to the possibility of Tintri not being long for this world, they have not "failed"….yet

*Tintri=Fat lady just started singing her first verse.*

- Nimble Storage is a ground up design that moves all performance into the compute layer, and as such is completely storage agnostic as well. Their very first product was indeed (in 2009 by the way) an "all flash array", however, due to their media agnostic nature, they pivoted to a hybrid array which was a MUCH larger addressable market. While they do indeed offer an All Flash Array, they have certainly not abandoned their hybrid array either. Oh, and if being purchased for $1.2 billion dollars is "failing", I can guarantee there a lots of folks out there that would LOVE to "fail" that way!

*Being sold for less than your IPO and 80% less than your highest market capitalization does not appear in the dictionary as definition of success. Nimble failed, first and foremost, by trying to take a product built for SMB and sell to Large Enterprise. That failure occurred October 2015 and admitted to by their leadership on their earnings calls. The creation of an AFA was simply a join the herd mentality. Without a path to profit, a sale became inevitable. By the way, Pure storage is next.*

Infinidat is a "me to" hybrid platform with large scale that wows potential customers with its 1PB marketing. Very capable storage, very large scale, but the performance is certainly NOT, nor can it ever be (unless RAM becomes MUCH less expensive) comparable to an All Flash Array.

Plain and simple, folks that can address ALL THREE requirements: Consistent Low Latency (Flash), Very Low Latency (Hybrid), and Extreme Scale (PB+) are the overall companies that will "win”.

*This is precisely what Infinidat does. We back it up with our Faster than Flash Challenge. Furthermore, AFA != consistent low latency. We see in the field time and time again, response times higher than 5ms due to garbage collection, high writes, not to mention high latency with a controller failure.*

Right now Infinidat is addressing an available segment of the market that is getting less and less viable. As folks continue to tire of having silo's of storage (one for extreme capacity, one for low latency, etc) then they will have to make a decision that will be much more challenging to address in their current iteration...and putting 1PB of flash on a datacenter floor and not charging for it until the customer needs it is going to be VERY expensive!

*Our COD model is popular with customers due to its transparency and financial flexibility. Customers know exactly how much it costs them upfront and what it costs to consume more capacity. All without the requirement to consume beyond the initial amount. To many, such a model may seem expensive, however, in the decades that persistent data storage has existed, the demand for capacity has NEVER gone down.*

Sometimes we need a magnifying glass to understand what real growth is: If I have 100 customers and next year I have 200 customers, that is a 100% growth pattern, and so on...but I am still irrelevant in the overall market until I have thousands and grow by thousands!

Should any of the above "failures" have stuck to their hybrids? The answer is "yes", and by the way...that is what ALL OF THEM has done! FUD is not, and has never been FACT!

Just my $.02

[El Storage Guy]

Re: Tegile and Nimble aren't "really" AFAs: they were hybrids like Infinidat until recently

To AC, - INFINIDAT Employee here.

YOU SAID: "Infinidat is a "me to" hybrid platform with large scale that wows potential customers with its 1PB marketing. Very capable storage, very large scale, but the performance is certainly NOT, nor can it ever be (unless RAM becomes MUCH less expensive) comparable to an All Flash Array."

>> This is what happens when platforms are compared based on Spec Sheets. HW is the defining element of traditional, tier-based, RAID-based architectures. Traditional Hybrid-Architectures are NOT truly smart when it comes to data properties and that is reflected on their sub-par Cache Hit-Ratio's. (which means you 100% depend on Disc/Media speeds). InfiniBox does not rely on media... as a matter of fact, we don't care whether is SAS or FC spinning rust. What matters is: We get the SPEED, We lower the cost, and we deliver 100x more reliable storage. To me, that's what customers really want.

Note: I personally had tested and replaced VMAX AFAs, XtremIO, Pure, Solid-Fire, V9000's and A9000s, Violin and I'm sure I'm forgetting others. I had been part of the Performance Engineering teams and ALL Flash teams of my previous employers (EMC and IBM).

YOU SAID: Plain and simple, folks that can address ALL THREE requirements: Consistent Low Latency (Flash), Very Low Latency (Hybrid), and Extreme Scale (PB+) are the overall companies that will "win".

>> I honestly believe that is an INCOMPLETE statement. It has an engineering perspective, but in the end, customers don't care if it is SSD, All-Flash, Hybrid or not. (They should not), They (should) care about price, reliability, speed, latency, ease of use, space, integration, support, ROI, stability and future of the company (Pure is dripping money every day), future-proof and overall: value. None of that should be dependent on media (like SSDs and defined as AFAs).

NOW, If a solution can provide that to any customer, what's wrong with that?

[Anonymous Coward]

Re: Tegile and Nimble aren't "really" AFAs: they were hybrids like Infinidat until recently

Compellent has been doing the hybrid thing since 2003, with fast performance tiers and cheaper capacity tiers. Jump forward to the flash era and the compellent adapted very easily to this architecture, with flash and large capacity disk.

Dell bought Compellent. The product is now the SC series and according to the specs, this scales up to 3PB, has dedup, compression, block and file - and really good support.

In comparing the two, I don't see Infinidat as anything special or unique.

[Anonymous Coward]

Re: Tegile and Nimble aren't "really" AFAs: they were hybrids like Infinidat until recently

Infinidat isn't tiered storage. That's the big difference. The problem is the terminology. Everything which has flash and HDDs in the same box is "hybrid", no matter how it uses those resources.

Tiered storage is on its way out, mainly because there are essentially only two tiers now: flash and large capacity drives. It's inefficient because you have to keep moving data around and applications aren't kind enough to put all the hot data at one end of the volume and the cold data at the other end.

The future viability is dependent on whether the cost of flash will come down enough to make it viable for storing cold data. That's a long way off which is why the big vendors offer multi-product solutions, or tiered storage.

[Anonymous Coward]

Re: Tegile and Nimble aren't "really" AFAs: they were hybrids like Infinidat until recently

Infinidat may well be around longer than Compellent now it's part of the EMC portfolio and has Unity to compete with internally.

[Anonymous Coward]

Re: Tegile and Nimble aren't "really" AFAs: they were hybrids like Infinidat until recently

It is interesting that the 3 companies you mentioned pivoted to all-flash. To answer your question whether they should have stuck to hybrid rather than moving to all-flash - I guess it depends. At the end of the day, it's the job of the storage companies to deliver what customers want. Every customer wants reliability, performance, scale/density, simplicity/ease of use and lowest cost/TCO. That's what matters. It's on the storage companies to design architectures that deliver those 5 things to the enterprise. Some architectures require all-flash to deliver on those customer requirements because they can't get performance or simplicity without it....INFINIDAT is a software centric design and delivers across all of those requirements to the highest standard and because it's able to leverage whatever is the lowest cost media for the persistent storage layer (today and for the foreseeable future that's NL-SAS) without compromising performance, reliability, simplicity or density it's going to deliver a TCO that the all-flash array architecture can't touch.

[mikeymac]

Go Infinidat!

I'm bullish on Infinidat.

We have two of their F6000 systems, and couldn't be more pleased. Cheap, fast, uber-reliable and the best customer support experience I've ever had. It's been a game-changer for us.

My only fear is opening El Reg some Wednesday morning and reading something like: "HPE Gobbles Up Dense Disk Dealer Infinidat for 2.2B". *chills*

[zbmwzm3]

Re: Go Infinidat!

What was your cost per TB?

[mikeymac]

Re: Go Infinidat!

@zbmwzm3 - Unfortunately, I don't think I'm permitted to say. I always tell the Infinidat guys to brag about the price, but, for reasons that I can't divine, they seem reluctant to do so. You should have one quoted for you. You'll be gobsmacked! When you take price, performance, reliability and customer support into account, you simply can't beat Infinidat.

On a 1PB F6230, we're averaging <1ms on reads, and ~.35ms on writes. I don't know if we'd see better performance from an AFA from one of the MegaCorps, but it's more than fast enough for my use case, and I KNOW it's cheaper than the other options.

Good luck!

[briancarmody]

Re: Go Infinidat!

brian from Infinidat here. Thanks for being a customer. Have no fear, acquisition is out of the question. In fact it is more likely that Infinidat would acquire certain remnants of the incumbants in the coming years than vice versa.

[mikeymac]

Re: Go Infinidat!

Thanks for the reassurance, Brian! We're working on procuring a pair of F4000 systems for backup and recovery, so my fear of your acquisition hasn't stopped us from moving forward. ;)

[Anonymous Coward]

I think folks have a tendency of associating their few wins with overall success in the market.

I have seen Infinidat exactly 2 times in my area, talked extensively to the customers that have the storage and they love it! No reason to bash it, no reason to call it a bad product.

However, I have been to greater than 300 customers in the EXACT same area that are running Nimble (one of which is of the above couple running Infinidat by the way, but needed an ultra-low latency platform so they bought a Nimble AFA).

The number of Pure in the same area is about a fifth of the Nimble, Dell/EMC and NetApp is neck and neck.

The point above about Infinidat caring less about the media is EXACTLY the same for multiple storage vendors due to the fact that they have moved performance out of the spindle/media world. Nimble is a great example of this: No matter how many spindles you have...your performance doesn't increase.

Another valid point is the fact that providing consistent low latency, low latency, and PB scale in the same product is valid. The media should be irrelevant at this point, we have moved past that!

I don't see anyone buying Infinidat in the near future...not because they are a bad product, but because there isn't anything to buy!

[mikeymac]

You say: "I have been to greater than 300 customers in the EXACT same area that are running Nimble (one of which is of the above couple running Infinidat by the way, but needed an ultra-low latency platform so they bought a Nimble AFA)."

As I look through this thread, I think I'm the only contributor who actually owns Infinidat systems. I can assure you that we did not buy Nimble (though their analytics are pretty awesome!). :D Apologies if I misunderstood your post!

[Anonymous Coward]

Customer review - storage architect perspective

Capacity per Floor Tile is a fair argument, but I think that's about where it ends. What's the cost per TB that you smash onto that tile?

I've migrated 2 VNXs (7600 and 5800) and 4 XtremIO arrays onto a single F6000, and can attest to the fact that it was, indeed, faster than my all flash arrays (obviously beating the VNXs as well) for almost all workloads. XtremIO couldn't keep up with SQL Server 64k writes - write latency of 300+ms. My average write latency on the F6000 is about 2ms, and average read latency is 5ms. My max write latency in the past year was 5ms, and my max read latency was 12ms. This was a FULL array with the smallest cache option (24TB) - I'd expect better read numbers if I'd opted for a larger cache.

People ask about cost - I can't tell you specifics, but I can give you ratios. Compared to my OVERALL EMC average cost per TB, InfiniDat was half the price. When compression was enabled, I was realizing 2:1 compression on ingress data, moving my price to 1/4 that of EMC.

If you compare the InfiniBox pricing to XtremIO (even granting XtremIO a generous 6:1 storage reduction with compression and dedupe), it was still 2.5x cheaper.

During other RFPs, InfiniDat also beat NetApp FAS arrays, XIV, and Pure, both from a technical specs perspective and on the cost/TB front.

All flash arrays almost guarantee your bottleneck won't be the disks, but think about the controllers you're putting in front of them...if your disk can push more IO than your controller, you're wasting money.

As others have said, their support should be the standard for other companies to follow. Array firmware upgrades take about an hour, and are simple to schedule. If I have a problem, I have a direct line to a person assigned to my account (THAT I DIDN'T HAVE TO PAY EXTRA FOR *COUGH*EMC*COUGH*). The system is so redundant that unless I pitch a fit, they don't bother replacing failed drives until 6 of the 480 capacity drives have died.

[mikeymac]

Re: Customer review - storage architect perspective

"People ask about cost - I can't tell you specifics, but I can give you ratios. Compared to my OVERALL EMC average cost per TB, InfiniDat was half the price. When compression was enabled, I was realizing 2:1 compression on ingress data, moving my price to 1/4 that of EMC."

This is the big story here. PRICE it. POC it. PURCHASE it.

[zbmwzm3]

Re: Customer review - storage architect perspective

Anyone near $150 per TB yet? I've actually had some bids hit around $200/TB, but it lacked some nice features.

[mikeymac]

Re: Customer review - storage architect perspective

@zbmwzm3 - $200/TB piques my interest! (as long as it's not based on an overly optimistic data reduction ratio). ;-)

[Anonymous Coward]

Another Customer Review

We are early adopters of the Infinidat F6000 series arrays. While I don't have any AFAs to compare against I can vouch for the sub-millisecond response times with mixed workloads. Our version 1.x arrays didn't have any flash tier and we were still 1-2ms under heavy load. That is amazing considering it only has NL-SAS drives to service the IO. Please stop with the "Hybrid" comparison especially if you never tried it.

Code updates are a non event. Drive sparing is ridiculously fast. Interface is like nothing on the market. Surprisingly the 4 year support renewal isn't forcing us to buy new arrays or swap out our controllers. We now have 6 of them on the data center floor and growing. Going with Infinidat has made our lives easier.

[briancarmody]

Re: Another Customer Review

brian from Infinidat here.

First, thanks for being a customer. If you're an early adopter, that means you took an extra leap of faith and put a huge amount of trust in our team. Everybody knows how high the stakes are in data storage and that folks are putting their careers on the line when they make the call. I don't know who you are (and I assume you won't doxx yourself here), but thank you; I cannot emphasize how seriously the team takes this covenant with customers.

>Going with Infinidat has made our lives easier.

That makes me really happy. I cut and pasted your comment in a note to the engineering team, and no doubt you made everybody stand a little taller (or at least slouch less in their chairs) this evening. Thank you for sharing!

[Anonymous Coward]

Welcome to 2011

Other than creative marketing and fancy bezels, I don't see what attracts investors to a company that's building monolithic hybrid storage arrays and now virtual tape libraries?? There's a reason why other vendors are focused on all-flash and SDS.

[Anonymous Coward]

Re: Welcome to 2011

>There's a reason why other vendors are focused on all-flash and SDS

Because they can't think of anything better?

[Anonymous Coward]

Huge Fan of all things Storage related - The Infinidat story is a great one to follow.

To add to the positive comments here, Infinidat is also a mixed protocol play ( BLOCK and NFS ) - Capacity, features, price - all very attractive. I have been following Infinidat for about 3+ years and the big picture just keeps getting better. Look at previous ElReg articles on Infinidat and the related comment sections - the detail and usefulness of the comments are better and better. Again, big fan of all things storage but solid companies like NetApp and Infinidat and others, which push the envelope, just keep moving forward. Dive into the details and you will see they are pretty advanced, ( COD , triple controllers, mirrored writes, built in APC, massive data coalescing, wild algo's , block storage, cloud capable ) - recently heard a pitch on the recent Symetrix / Vmax - it sounded a lot like Infinidat - well the core Infinidat team "created" EMC so that makes sense -- now if only they can offer CIFS / SMB - Hmm ??

[briancarmody]

Re: Huge Fan of all things Storage related - The Infinidat story is a great one to follow.

Brian from infinidat here

Thanks for your comment. We feel acutely the lack of SMB support and are working on a pretty cool solution. It's taking a bit longer than expected because the protocol is actually way more complex than it seems on the surface, and we concluded that we had to write our own SMB server in order to deliver deterministic sub-ms latency with seven nines reliability. Let's just say I have even more respect for Samba creator Andrew Tridgell now than when we worked together at IBM, and the protocols were his problem :-)