Network-Attached Processing: One Powerful Idea

What if instead of adding servers every time you needed more computing horsepower for application development projects, you could plug in a network-attached appliance — to supply extra processing power and memory on demand? That’s exactly what Azul Systems promises with its Vega 2 “compute appliances”, the second generation of which debuted this week. These appliances clue in when you’re doing app development work in a run-time environment like Java, and take over the processing chores. All the Java work gets automatically offloaded from the server to the Azul box.

Put another way, Azul CEO Stephen DeWitt says you can think of what his company offers as processing power and memory as a service (no, let’s not create an acronym for that.) Or, you can think of its as processing power as a utility.

Think about that. An Azul appliance lets you augment your existing servers’ processing and memory power by plugging in an appliance, much the same as you already can add network-attached storage. You mount this as an external resource to whatever kind of servers you’re using — Lintel, Wintel, whatever. The Azul appliance doesn’t care what server platform you have.

Whether you have a bunch of boxes on your network strung together or you have big iron, you see the possibilities for saving money and space. Most of all, capacity planning becomes a lot less of a white-knuckles exercise.

“We have separated the functionality away from the server and put it in the cloud,” DeWitt says.

The immediate appeal of Azul’s approach is the flexibility. At a time when businesses increasingly demand that CIOs turn app development projects around quickly to support business side innovation, an appliance like Azul’s could provide peace of mind – since it lets you add tremendous horsepower to app dev projects as needed, while ensuring the rest of the business-critical apps built in Java continue to run smoothly.

This isn’t measly processing horsepower we’re talking about, either. Azul’s first-generation products packed up to 16 Vega chips, each one with 24 processing cores, into one appliance. (Azul designs the chips itself and uses a partner to fabricate them.) The second-generation appliances Vega 2 chips each carry 48 processing cores. Put 16 of those Vega 2 chips into one appliance (the upper limit) and you’ve got 768 processing cores at your applications’ disposal.  Plus an improved instruction set optimized specifically for Java (unlike say, Intel and AMD instruction sets, which are optimized for general PC use.) As for memory, Azul gives each appliance up to 768GB.

In addition to your in-house app development projects, the appliance will handle any other apps built in Java, say apps from BEA or Oracle. Basically, if it’s Java, it moves over to the appliance.

Banking is the biggest vertical market for Azul to date, but other high-transaction type environments also like the implications for scale. (British Telcom is using the appliances to build in processing capacity on demand for telecom companies who now must be provided with access to BT’s network due to telecommunications deregulation in the U.K.)

People are serious about consolidation of boxes in data centers, DeWitt says. True. CIOs also want better-than-Dell type economics and low-power footprints. Right again. Could network-attached processing help you achieve these goals? Let’s hear what you think.