Computing Becomes a Matter of State - Systems Management News On The Web

CHANNELS

ON THE WEB

PRINT EDITION

BZ MEDIA

AS OF 8/20/2008 9:18AM EST

Computing Becomes a Matter of State
By Alex Handy

August 5, 2008 — The move is on to banish state.

At the LinuxWorld Conference and Expo in San Francisco today, Jeffrey Birnbaum, Merrill Lynch's chief technology architect and managing director, spoke about his efforts to move the financial services firm toward a world in which the state of every application, desktop and server is stored in one place. Doing so, he explained, would eliminate many of the headaches enterprise IT organizations have had to suffer.

Birnbaum called stateless computing something of a misnomer. “Stateless computing isn't about having no state; it's about where that state is stored,” he said. “Since the 1980s, we've stored the state on the computer. That's presented the myriad challenges that have sent costs through the roof. The basic idea of stateless is that we want to move from a dedicated server network to a shared server network.”

Today, Birnbaum said, “the typical businessperson comes in and says, 'I have an application I need to run, and I need 20 machines.' What we want to get to is, 'I need x amount of capacity, and I need it for this amount of time. I need 20 units of capacity, and at the end of the month I need 100 units of capacity.' In some sense, that's the bigger definition of a cloud. Amazon is offering cloud services, where you need x amount of capacity.”

That's quite similar to where we were 30 years ago, said Birnbaum. Like virtualization, stateless computing has its roots in old mainframes. Essentially, it's a less restricted form of time sharing, where users push processes into the queue, and the machine dictates what gets done and when.

At the center of Birnbaum's plan is the central enterprise file system, into which all configuration data for all applications and operating systems are kept, and the placement server that manages the distribution of all work and application loads. Among the benefits of maintaining state on a shared network are configuration management, policy enforcement and auditing, rapid provisioning, and software upgrades and deployment.

The file system currently exists, but the placement software isn't there yet, said Birnbaum. He expressed hope that the industry will come together to build a complete solution, adding that some of the work Merrill Lynch is doing in this area may be open-sourced.

As for the software that will run on those stateless clouds, Birnbaum suggested that Google has the right model. That company, he said, standardized on unreliable commodity x86 hardware, then wrote its software to accommodate that lack of stability. He said that this offers far more flexibility and significant savings over buying blade-based systems with massive redundancy built in.

While Birnbaum's approach may increase the complexity of the software, he said stateless computing offers enough of a savings on time to make developers happy.

“When you start with this idea of an enterprise file system, you are making it very easy for developers to distribute their code and not have to worry about the differences between all these systems,” he said. “The beautiful thing about the enterprise file system is when you refer to a version, it's the same version everywhere. The placement engine should be smart enough to say, 'I'm not placing an application; I'm placing a system.' It has a manifest that says, 'I have to start A, then B, then C,' and it should handle that.”

Related Search Term(s): grid computing, Linux, Google

EMAIL THIS ARTICLE

SEND FEEDBACK