Philippe Nicolas (PN): Mike, recently The IT Press Tour met up with iXsystems. Could you share with us the genesis of the company, as your products are much more known than the company behind (them)?
Mike Lauth (ML): My partner Matt Olander and I acquired the hardware business of Unix vendor BSDi in the early 2000s when the company was being divided up and sold off. The software side of the business went to embedded software vendor Wind River Systems. We think we got the better end of the deal because the hardware manufacturing operation not only had strong capabilities but also a valuable portfolio of Silicon Valley customers. We knew that if we treated them right and watched our bottom line, we could ride every wave and weather every storm that tore through the Valley. And we have!
PN: You're a true believer of Open Source, what are the advantages you see in such philosophy for you as a builder and for users?
ML: We have always been in the server hardware business and have never believed in putting artificial constraints on the hardware we sold to limit or control how it is used. That may sound obvious for hardware but the same philosophy can also apply to software. Open Source software should never have any artificial restrictions placed on it that might limit what you can do with it. In practice, this makes open source software an equally-solid building block as the hardware to a business that chooses to embrace it. While permissively-licensed software like FreeBSD has been criticized for not forcing companies to “give back”, few companies actually survive the closing of software that is otherwise open to everyone. We follow the FreeBSD model with FreeNAS and at over ten million downloads, we’ve learned that giving away our storage software with no strings attached can win you millions of users, thousands of quality assurance partners, and many thousands of hardware customers. With FreeNAS we leverage the community as part of the QA process; users get a product that has been exposed to thousands of QA partners in exchange for meeting their storage needs. The availability of the FreeNAS source code allows members of the community to change the software to meet specific needs and this transparency also allows a business to ensure the security of their data.
PN: What are the open source projects you drive or ones in which you are a key participator?
ML: We develop the TrueOS desktop/server operating system based on FreeBSD but our runaway hit is FreeNAS, the Open Source file, block and object enterprise-grade Open Source Software Defined Storage. The FreeNAS project was facing an uncertain future back in 2009 and as FreeNAS users and community members ourselves, we didn’t want to see the project fade away. We negotiated to acquire the FreeNAS project in 2010 and set about rewriting it, modernizing it, and incorporating the ZFS file system. Multi-terabyte hard drives had recently gained adoption in the market, and we instantly recognized the potential of FreeNAS to deliver enterprise-class storage services, coupled with our hardware, to a market that was desperate to leverage Open Source economics to address rapid data growth.
PN: You produce and build your own products with your assembly entirely in San Jose? Please explain to us the production chain, what do you source outside…? From whom…?
ML: We are definitely not a VAR; we are a System Builder. We don’t resell off-the-shelf systems like many other vendors do. Instead, we source individual components and platforms from leading distributors and ODMs for assembly, integration, and testing at our San Jose facility. Even though we have a standard set of products, most systems we produce can be custom-configured. Our goal is to be flexible and build our products around our customers’ requirements instead of requiring our customers to squeeze their requirements into a narrow product line, as most of our competitors do. You name a manufacturer or distributor and we buy from them, giving us a global footprint that rivals our largest competitors.
PN: Also, do you serve other HW vendors? If yes, who are they and what are the reasons they select iXsystems?
ML: In addition to creating custom internal SKUs, as well as our our own TrueNAS and FreeNAS storage appliances, we do build server appliances for partners who brand them to meet their needs. There is a huge gap between buying off-the-shelf servers and ordering container loads of custom systems from Asia. We fill that gap well by combining flexibility with raw on-site assembly capacity. Of the partner relationships we can discuss openly, we build the ScaleEngine media streaming appliance which, like Netflix’s Open Connect appliance, gets positioned at strategic data centers around the world to service regional loads. Keeping their server and storage configurations standard is critical for ScaleEngine to manage and scale their operation.
PN: How many customers do you have? How many systems do you produce a month? May we have the split between servers and storage? How many PB did you ship in 2017?
ML: We have thousands of customers around the world, many of which have benefited from a multi-decade relationship with iXsystems. Our customer base runs the gamut from hyperscale data centers and customers building public clouds to individuals deploying our FreeNAS Mini storage system in small offices. We called 2016 the “Year of the Petabyte” due to the sheer number of customers, such as McGill University we earned that were using TrueNAS to manage multiple petabytes of storage. By the end of 2017 we had nearly 5,000 customers and over 1 exabyte shipped. Both our server and storage businesses continue to grow year over year, but we do see faster acceleration in our storage growth. This is due to the increased reach of companies using FreeNAS. It’s pretty common for us to talk to one of our server customers and end up finding out that they’re also using FreeNAS in some capacity for their storage needs. Those customers are pretty excited to learn that there is an enterprise version of FreeNAS, which makes the transition to TrueNAS seamless for their mission-critical use cases. Those customers typically have experience using FreeNAS and love it but need a supported storage solution for business operations, which makes for an easy sale and migration. We’re always going to continue to serve the enterprise server market, but it is really exciting that we have made a name for ourselves as an enterprise storage company over the past seven years.
PN: I understand you’re profitable, what is your current growth rate? In Servers and Storage?
ML: Server sales have been solid for decades and regularly experience surges with new data center deployments. Storage sales have consistently seen over 50% annual revenue growth since launching TrueNAS in 2011 and are taking on a life of their own.
PN: iXsystems acquired and then rewrote FreeNAS. What were the reasons behind this enormous task?
ML: When we acquired FreeNAS, its GUI was rather dated and it was lacking important enterprise features. We knew that ZFS was a natural addition to bring those features into FreeNAS and so we concluded that a ground-up rewrite was due. The original FreeNAS PHP-based GUI was limiting, so we moved first to a Django-based web framework which we are now retiring in favor of an AngularJS-based one later this year. The initial rewrite was indeed a monumental task but we’ve found that it addressed many customer needs. The rewrite also allowed us to move to a more evolutionary development model thanks to our separation of the FreeNAS middleware from its user interface. This also allows us to provide a “REST” API for automated management of FreeNAS systems.
PN: FreeNAS is clearly a best-seller if I can use that term. Could you tell us which partners use it and how many installations you have so far? What is the total capacity FreeNAS has under control?
ML: At over ten million FreeNAS downloads we have put OpenZFS in more hands than anyone and we believe this is true to the vision of the original ZFS team at Sun. We have indeed seen other vendors around the world either ship hardware tailored to FreeNAS or offer a turn-key solution with it including video editing solutions in Japan, Scandinavia, Canada and the USA. These are not official partners but we still see their efforts as extending our marketing reach by increasing our installed base. In addition to countless FreeNAS systems hiding behind corporate firewalls, we know that over 200,000+ FreeNAS systems check for updates on a regular basis. The exact number of FreeNAS systems will never be known but we conservatively estimate it to be at over half a million. As for the total number of Exabytes under control of FreeNAS, your guess is as good as ours given that we do not spy on our users. If every one of our more than 200K FreeNAS users has 4TB under management -- again a very conservative estimate -- this would mean that FreeNAS users around the world manage in excess of 8 exabytes.
PN: What was the motivation also to incorporate OpenZFS?
ML: OpenZFS truly is as good as people say it is. It is a next generation file system without equal and nothing can touch it at any price, nor will they for some time. It encompasses the functionality of traditional file systems, volume managers, RAID controllers, and more, with consistent reliability, functionality and performance. From its continuous data validation to its unlimited snapshots, OpenZFS truly puts the competition to shame.
PN: Now about TrueNAS, the enterprise flavor of FreeNAS, what are the differences? What do you bring to the table? What about scale-out storage?
ML: It’s funny but we give away the number one Software-Defined Storage solution [FreeNAS] while others struggle to make a profit from SDS. We learned the hard way that enterprise hardware and software must be tightly-integrated to provide enterprise-class performance, reliability, and a consistent customer experience, and TrueNAS is exactly that: a custom, user-serviceable hardware platform with enclosure management, high-availability, performance optimizations, 3rd party certifications, and up to 24/7 global support. I look at an Android phone as a perfect example of how the availability of the software does little to provide you a meaningful experience as a user until you seamlessly match it with custom hardware. And, who makes the best Android phones? Arguably, Google. Why? Because they can design the software and hardware in concert. As for scale-out, we are very pleased with OpenZFS’ scale-up architecture because it can not only support multiple petabytes with today’s hardware but also provides the building block of a yet-larger scale-out infrastructure. Unless it’s part of a truly replicated infrastructure, scale-out is more often a high-latency crutch for limited-capacity storage systems than truly scale-out storage. Scale-out storage has its place, but when you can build scale-up architectures in double-digit petabyte scale that can outperform most scale-out storage for fractions of the cost, you start to question its importance.
PN: I understand that the TrueNAS M-Series represents the third generation of TrueNAS. What features distinguish it from previous generations of TrueNAS and competing products?
ML: While some upgrades are no-brainers like upgrading from SAS2 to SAS3, doubling max capacity to 10.5PB while decreasing the physical footprint, and enhancing throughput to 100Gb ethernet and 32Gb Fiber Channel, we have really embraced flash technologies like NVMe and NVDIMM with the ”M” Series platform to achieve the best price/performance ratio possible. Any vendor can replace their HDDs with SSDs for a performance boost but integrating NVDIMM technology requires careful coordination between software and hardware engineering. Technologies like NVDIMM truly separate even the highest-end DIY FreeNAS systems from TrueNAS.
PN: You offer block and file storage, what about object storage such as S3? Any plan and what is/who is the one you have selected?
ML: Object has indeed joined file and block as the third essential network storage protocol and we are meeting demand for it by both serving S3-compatible storage and by supporting replication to Amazon S3, Azure Blob Storage, Backblaze B2, and Google Cloud Storage. True to our open architecture, we would like to support every popular cloud service available.
PN: Is there a lot of upsell between people who start with FreeNAS and then adopt TrueNAS or do you sell TrueNAS directly? In that case is it to replace some brands? And what are the top 3 brands you replace?
ML: Given that we have never met the majority of FreeNAS users, most of them work their way up our storage product line on their own, driven mostly by word of mouth. FreeNAS isn’t a household name, but it is a household name in IT. So, we focus our marketing efforts on TrueNAS, but we cannot ignore the steady flow of users who discover FreeNAS at home or in their dorm room and bring their storage management skills to work. I’d say that Dell/EMC, NetApp, and HPE are our top competitors, all of which we aggressively challenge on value -- measured by features, performance, and capacity per dollar -- and with our streamlined storage product line.
PN: Can a general-purpose hybrid and AFA solution take on niche AFA solutions?
ML: I would argue that they do every day but it comes down to the individual workload that a user is throwing at their NAS system. While an AFA may be needed for a specific workload, it is a very expensive “easy button” for a mixed “unified” workload. We encourage users to carefully analyze the characteristics of their real-world workloads and understand when AFA or hybrid are the way to go, or even have a clean division of the load between two systems. Your database may demand AFA while its backup is obviously fine with only rotational media. Niche solutions are exciting but we’re still focusing on the careful pairing of a general-purpose system to a workload because a lot can be achieved by understanding the requirements and abilities of each.
PN: What storage trends do you think will drive the next five years?
ML: There is no question that persistent memory -- like NVDIMM -- is a perfect glimpse at the future of many aspects of the storage stack: silica sits adjacent to silica with few busses or abstractions to slow it down. “All NVDIMM” storage however will only serve the smallest of workloads for years to come, leaving NVDIMMs as useful components in a larger storage stack rather than as a platform for primary storage. With the possible end of Moore’s Law rapidly approaching, the optimization of the storage technologies we have today will play a key role in the years to come. Brute-force solutions like AFA-everywhere will not be economically viable for some time. I also feel that as exponential data growth continues, the need for Open Source economics becomes all the more pressing. Addressing that need just isn’t sustainable for most companies with the “traditional” vendors or the public cloud.
PN: Do you have a hyperconvergence play and if so, how will you distinguish yourselves from niche players like Nutanix?
ML: The recent Meltdown and Spectre CPU-level vulnerabilities kindly reminded us that as with flash technologies, virtualization is a useful complement in a larger computing stack rather than its sum. Our hyperconvergence play began nearly a decade ago with “Jail” containers in FreeNAS and now the addition of the bhyve hypervisor. We are very excited about the potential of these technologies but have not yet integrated them into TrueNAS given the surprises that hosting an OS within an OS can introduce. You should not jeopardize the performance of your storage system for the sake of saving on an attached compute device. Careful capacity planning aside, the real question is how hypervisors can serve the storage stack. For example, a consistency-sensitive backup scenario such as a database is not adequately validated simply by knowing that you have a duplicate of its raw data. To be properly validated, a backup of that database should at a minimum be imported into an on-backup database server and ideally be interrogated by a validation application. The online nature of OpenZFS unlocks opportunities like this that were never possible with tape. You really want to be the person who can confidently say, “Yes boss, our tertiary backups are good” when asked, and a combination of OpenZFS and a hypervisor will ultimately achieve that.
0 commentaires:
Post a Comment