When I first heard about Infinidat I put it in the category of FM — “Effing Magic.”
At the time, Wikibon analyst David Floyer was shaping the industry’s first prediction of the so-called All-Flash Datacenter. Flash prices were falling faster than those of spinning disk (we think they still are, by the way), and the writing was on the wall for the traditional disk array. As anyone close to the business knows, Infinidat is led by tech visionary Moshe Yanai, who along with Dick Egan changed the storage business permanently in the late 1980s by bucking the generally accepted way of doing things.
Moshe and his team are at it again.
In classic Moshe fashion, he convinced a group of rock star engineers in his network to join him. Most of these individuals probably don’t have to work but do so because they like to solve gnarly problems. Infinidat uses intelligent algorithms to squeeze the most out of storage media, including spinning disk. There are very few storage companies – if any – that see the world the same way. Infinidat is making some bets that if correct will give them a leg up on virtually all their competitors. The two biggest of these are: 1) that the rotating rust called hard disk drives will continue to be substantially less expensive than flash, indefinitely and 2) by using math, the company can squeeze out comparable and sometimes better performance than all-flash systems.
At Wikibon, we’ve had some spirited conversations with Infinidat folks, including Moshe, about these assumptions. We disagree with the first – Wikibon projections show that flash prices will continue to fall faster than those of spinning disk and will eventually crossover in all sectors. Way back in 2008, Floyer predicted that flash costs would drop below those of 15K HDD by mid this decade and he’s continuing that projection into mainstream devices.
Floyer’s projections have been quite accurate, but recent research, prompted by the Infinidat discussions led Floyer to investigate further and we’ve had deeper interactions with Seagate, Western Digital and others to test these assumptions. We’ve also taken a hard look at the projections by TRENDFOCUS and old friend Mark Geenen and his team. It has led to a change in our thinking with respect to enterprise-class HDD and Floyer has elongated the crossover time for these devices. Nonetheless we still expect a crossover to occur even in this space and will shortly publish these projections.
The cost/price projections tend to be very nuanced. Raw vs. compressed/de-duped, device cost vs. TCO, the impact of data sharing and many other factors. Floyer will go all day, toe-to-toe with anyone, debating these trends but we always keep an open mind and are happy to change our views if the research points us in a new direction. The bottom line of my many discussions with Floyer on this is we still think flash wins out by the end of the decade.
Here’s the thing. If Wikibon projections are wrong, or even too aggressive, Infinidat gains a considerable advantage over other hybrid competitors because its costs will be lower. If we’re correct and flash swamps the spinning disk industry, for Infinidat it doesn’t matter because the company can thrive in an all-flash world. It’s magic works on spinning disk or flash.
Rapid & Innovative Product Cycles Win in Storage
The storage companies that have been successful are those that can turn R&D into product and have a constant cadence of innovation and improvement moving from lab to market. Indeed, many of IBM’s storage challenges, for example, relate to the company’s lack of rapid product innovation. Up until recently the pace of innovation was too slow. EMC on the other hand, while often criticized for lack of innovation, was always able to get new products to market quickly and move the installed base forward.
Almost exactly one year from its 2.0 release (NFS and asynchronous replication), Infinidat today announced three significant innovations as part of its 3.0 release. Importantly, these innovations are delivered via software updates to any customer on active maintenance, back to serial number 00001:
- Support for inline compression. Infinidat’s goal according to Chief Technology Officer Brian Carmody, was to develop a compression algorithm that allows an increase in capacity with no latency impact. The implementation Infinidat uses is that when a piece of data hits one of the three controller nodes inside of its InfiniBox system, it writes that data to memory on the controller node and then sends a copy via a DMA transfer over an Infiniband network to a second controller node’s memory. Once the data is copied over, the system acknowledges a write. This operation takes 180 microsecs. Infinidat implements the compression engine on the backend as an asynchronous process. When data needs to be persisted from memory, Infinidat pulls data sections from memory and assembles them into de-staged 64KB stripes and writes them to persistent storage, along with parity and data protection. When de-staging, Infinidat chooses from a library of algos and makes a “data aware” decision based on the most appropriate compression algorithm and applies it to each 64KB data chunk. By compressing data as an asynchronous operation, Infinidat avoids the overhead typical of all compression implementations. This means Infinidat can apply in-line compression to spinning disk without a performance hit, further supporting its premise that spinning disk costs will remain lower than flash.
- Major update to its iSCSI implementation. iSCSI has been perceived by many storage practitioners as a second level storage protocol, designed for business, not mission critical workloads. While many experts (e.g. Jon Toigo) would entertain debates on this issue, FC bigots remain in force. Infinidat claims as part of 3.0 it is bringing iSCSI to carrier class service providers by making its iSCSI service a peer to FC and FICON to support world class storage services for customers wanting Ethernet. Infinidat Chief Marketing Officer Randy Arseneau calls this “a big boy implementation of iSCSI.”
- Performance analytics. Infinidat claims it has advanced modern performance analytics and tooling beyond conventional implementations by bringing multi-dimensional analysis with detailed granularity to its 3.0 release. The company allows customers to analyze storage system data and build OLAP cubes of (for example), individual SCSI commands, NFS operations, specific IO patterns, etc, allowing for very detailed drill downs and analytics. The company has developed a GUI with nice visualizations and also makes all the data available via a REST API if customers want to use other visualization tools to perform analysis (e.g. Splunk, Tableau, Qlik, etc.).
A Company to Watch
A so-called Unicorn (i.e. a startup with greater than a $1B valuation), Infinidat appears to be winning in the market. With 400 employees by year end, a half-exabyte of production capacity in the field worldwide and a consistent growth rate, the company appears to be on track for either an initial public offering of shares or a successful exit. Personally I hope the former. The successful Nutanix IPO this week gives hope to companies like Infinidat that a storage (or storage-oriented) company can go public and do well. Wikibon estimates that Infinidat revenues will blow through $100M sometime next year which sets it up for the next phase of its growth plan.
Here’s a video of an interview I did recently with Carmody and Arseneau:
Disclosure: Infinidat is a client of Wikibon, a division of SiliconANGLE Media, Inc., of which I am co-CEO. Opinions expressed are my own and those of Wikibon. Clients have no editorial control over our work.