Mating Grid And Cloud Software With Hardware

Like politics or economics, computing goes through cycles. Corporate computing started out highly centralized on mainframes and other room-size machines, but since the introduction of the PC has been going through swings between the autonomy of the individual computer user and centralized management by the IT department.

A conversation with Nati Shalom, chief technology officer of GigaSpaces, got me thinking about another pendulum swing: the one between making applications optimized for specific hardware vs. being independent of it.

Operating systems designed to run on hardware from multiple manufacturers and Java virtual machines promising portability across hardware and operating systems were part of the swing toward platform independence. That’s in contrast to mainframe applications, or even versions of the "open" Unix operating system that were specific to a given computer manufacturer’s hardware.

GigaSpaces, an Israeli company founded in 2000, creates software for data analysis and processing on grids of computers, where each server tackles one part of large problem. This is the way large web operations such as Google ( GOOG – news – people ) and Facebook manage Internet-scale data processing, although the Web players tend to take a do-it-yourself approach based on internally developed or open-source software. GigaSpaces provides a commercial alternative that has also won fans among some Web, cloud computing and online gaming firms, as well as financial services and telecom players with very large data analysis problems.

Last week GigaSpaces announced it had created a version of its product to target Cisco‘s ( CSCO – news – people ) Unified Computing System. (That’s Cisco’s attempt to provide a complete data center hardware lineup, with its own blade servers and storage systems in addition to the networking gear that ties them together.) I remember wondering what Cisco was thinking by getting into the server business, but Shalom sees the advantage of reimagining the server from a network-centric perspective.

Shalom says his firm’s decision to target specific hardware, in pursuit of higher performance and greater scalability, is in some ways similar to Oracle‘s ( ORCL – news – people ) recent moves to leverage the proprietary server hardware it got with its acquisition of Sun Microsystems ( JAVA – news – people ). In particular, Oracle’s new Exalogic Elastic Cloud is "very similar in concept, but different in implementation" compared with the combined offering from GigaSpaces and Cisco, says Shalom.

One of the fundamental challenges of the grids that GigaSpaces operates in is the management of a large pool of servers, the data processing jobs they are assigned, and the results they return. Shalom says one of the main attractions of Cisco UCS is that it comes with an application programming interface (API) that gives the GigaSpaces Extreme Application Platform (XAP) management console direct access to the hardware. That makes it possible to see which blade servers are available for use, automatically provision them to an application, and add or remove blades as needed, without going through the operating system or any other intermediary software.

Automating the setup and configuration of the grid eliminates a lot of manual tweaking, Shalom says. "We know from experience that a lot of that configuration and tuning cycle is very painful. But with this, everything is pre-baked and pre-tuned."

Secondly, Cisco Extended Memory technology allows each blade server in a cluster to support more than twice as much memory as a conventional two-socket server, which makes it easier to construct grids that process terabytes of data in memory. That’s significant because analyzing or manipulating data in memory is much faster than retrieving it from a hard disk.