GigaOM

On-demand computing promises two things. One, the ability to grow or shrink capacity based on need. And two, the ability to drag and drop virtual machines instead of racking and stacking physical ones.

With today’s on-demand services, although the machines are virtual, they still need to be defined and managed by real people, one server at a time. As a result, much of the expected savings from virtualization never really materializes. Elastra, a San Francisco-based startup backed by Hummer Winblad Venture Partners, wants to change this by letting IT teams provision entire application clusters of whatever software they choose, into any on-demand computing platform, automatically.

Here’s how it works: Elastra uses a pair of document formats that describe what the application does and how it does it. Feed both into the company’s “cloud server” and you get a virtual data center, complete with monitoring, accounting and automated deployment, running on the virtual infrastructure of your choice.

Fixing IT with markup languages has been done before. The Data Center Markup Language was one cross-vendor attempt to describe data center environments and the dependencies between components. But Elastra CEO Kirill Sheynkman says his company’s different: Because virtual machines are software, IT teams can not only use Elastra’s language to describe their application cluster, they can also “run” that description and provision the application in its entirety on whatever virtual infrastructure they want.

The company’s first release, launched today, provides a relational database atop Amazon’s EC2/S3 platforms. “We did database first because it’s challenging enough to be interesting, but you’re not solving the world,” Sheynkman said. Eventually, the company plans to support LDAP servers, databases, message buses, search engines and other application building blocks. Elastra could also pave the way for licensing and compliance of enterprise software running on demand. “Commercial software ISVs (independent software vendors) have a problem running in the cloud,” he said. “It’s fairly simple to put a license server in there.”

Elastra doesn’t care which virtual infrastructure you use: It wants to separate the underlying virtual computing infrastructure from the task of system design, deployment and administration. Some firms will want to run their own internal cloud, too, and the company has built its cloud server with that in mind. “Lots of big companies want to offer a public compute utility,” said Sheynkman, “but there are also private companies big enough to do it on their own terms.”

The industry is gradually moving from virtual machines to virtual application clusters. By turning a description of the cluster into running code on the computing infrastructure of your choice, companies like Elastra get IT one step closer to the true potential of on-demand computing.

If this story interests you, check out our upcoming conference:
Mobilize — Mobile Web Today and Tomorrow