Azul: A $14.4 Billion Server Market Gone Wrong

The evolution of computers has long been watched, nurtured, and eagerly awaited. So predictable is natural technological growth that many maintain a justified confidence that today's science fiction will be tomorrow's pricey toy, as Azul Systems CEO Steph

Stephen DeWitt: Good morning (or afternoon) everyone.

Mike Vizard: Hello and welcome to Ziff-Davis IT Link. Why dont you tell us a little about who you are and what Azul does?

DeWitt: Great to be here. Im Azuls president and CEO and we are based in Mountain View, California. Guess I left off my name ... Stephen DeWitt. Azul is the pioneer of network attached processing, a revolutionary new model for delivering processing power and memory as a shared service. Azuls computing appliance was built from the ground up for todays development realities. Our architecture delivers unique performance and scaling attributes to any and all applications built leveraging Java or the J2EE platform.

Vizard: So the topic at hand is about how the server market has gone wrong. Care to elaborate?

DeWitt: I think the reality here is that the legacy computing models are failing. Enterprises are seeking to build applications that bring new performance and scale characteristics to their businesses. In virtually every other tier in classical data-center thinking, core services (storage, networking, etc.) are delivered as a fabric or service. Except in compute. We are trying to change that.

Vizard: Those legacy models then, are they the way we buy and deploy servers today?

DeWitt: Sure. Every application deployed has performance/scale expectations around it.

Vizard: So what exactly are you proposing as an alternative model?

DeWitt: In traditional thinking, you either attempt to meet those objectives by clustering lots and lots of little boxes together in a one app, one cluster-type model, and hope youve sized it right, or you spend big money on big SMP systems. The good, bad and ugly associated with these models are well understood ... remember all of the hoopla about utility computing?

Vizard: Yes. Are you not proposing a variation on a utility computing model?

DeWitt: Most solutions tendered by the legacy vendors were simply orchestration layers above a bad model. We see it differently, somewhat. We believe in a future where capacity planning around the functions of compute and memory no longer requires precognitive thinking. If an app needs a ton of processing power and memory, it is there. If it only needs a little, the rest is fully available for other apps to share, and the provision of that is dynamic. Look at NAS. NAS changed the face of the storage industry.

Vizard: So, the concept is that you build an application for a server, but as it scales, you augment that with your system that provides a platform that allows me to dynamically scale the application. Is that about right?

DeWitt: We no longer provision storage through legacy approaches like NAS. Instead, all apps/users/servers leverage a common pool managed as a centralized service. We do the same with computing. Developers today build in Java (or other managed run-time environments). That app is then deployed on any host system ... Lintel, Wintel, Solaris/SPARC, AIX/Power, virtualized blades, whatever. Those host systems then mount the Azul compute pool as a transparent asset.

Vizard: Why does this only seem to apply to Java environments?

DeWitt: Just as your notebook can mount TBs of external storage off in the clouds, apps (services in a SOA world) can mount effectively unbound processing and memory resources ... built from the ground up for these architectures, because you can only do this in the world of application virtual machines.

Vizard: Why are the server vendors seemingly unable to replicate your approach?

DeWitt: Weve pioneered the capability for segmented VM. So a host machine can leverage power elsewhere transparently.

DeWitt: Ah, the $64 question. Well, our approach has a significant impact on the amount of servers youll need, i.e., a lot less. We also have an instruction set in our systems highly optimized for run-time environments. No garbage collection/locking overhead, etc., and that flies in the face of the traditional general purpose–type approach that has defined the systems world for years.

Vizard: Have you partnered with any of the server vendors or are they trying to pretend that you dont exist?

DeWitt: A shared computing model is inevitable ... just as shared storage and networking in general evolved ... this is just a matter of time before this becomes a standard approach. We are in the first couple of years of this transition.

DeWitt: Ha Ha ... I think a little of both. Weve got a good working relationship with most, as we are in some of the most sophisticated IT shops that exist. So, inevitably weve seen our gear mounted by just about everything under the sun ... no pun.

Vizard: Can you describe some of the customer deployments?

DeWitt: Sure. Ill stick mostly to the top-level. In financial services, we are in production at a number of banks running everything from core risk engines, to trading platforms, to compliance apps, to front office apps and portals. In telco, we are also doing some interesting things. BT, for example, has Azuls technology as the core computing engine for their Open Reach gateways. Those are the gateways between BTs network and the rest of the ISP/telco, etc., vendors in the UK. We also have a number of e-commerce, Web, insurance and logistics customers leveraging our gear.

Vizard: How hard is this to set up? Do I need 10 guys in lab coats?

DeWitt: One very exciting trend is that we are now seeing significant growth down market. We are not just for the big guys. Many Web 2.0-type firms are able to leverage our power to highly differentiate themselves from the incumbents. You may not be aware, but our pricing starts at under $50K, and that is for our 96-way computing platforms. No one has ever brought this power to this class of customer at these kinds of economics.

DeWitt: Hard to set up? No, its an appliance!

Vizard: How does the pricing compare to buying traditional servers?

DeWitt: Again, this is where a shared service model and the dedicated approach have vastly different economics. We deliver to our customers a compute pool. That pool (like TBs of NAS) has a capacity. Ones, or tens, or hundreds of apps can mount that pool. As it gets 40-, 50-, 60-percent full, just add more. In our experience, weve delivered significantly better ROI and significantly better TCO in the customer environments we are in. Most importantly, though, we are non-disruptive. No OS change, no binary baggage, etc. We mount transparently, quickly and eliminate major performance/scaling bottlenecks with virtually no effort.

Vizard: What do you see as the major factor holding back adoption of this model?

DeWitt: Just time ... and experience.

Vizard: No cultural issues? Seems to me there would be a fair amount of inertia there?

DeWitt: Like most disrupters, weve been focused on the validation of our technology. This is key for our customers and for us. Now, with a handful of quarters under our belt, we are well past the "who is Azul" and "does it work" stages. We are running some world-class apps in some world-class environments. We said from the get-go that was a key focus out of the gate. I would agree that culture would be a barrier if we forced customers to change. We dont.