NASA and Boeing launch grid systems to share resources, contain costs.
Piyush Mehrotra isn’t exaggerating when he says the data collected and maintained by his organization’s computing systems spans the universe.
In fact, data collected by instruments in space might overload even the most advanced computing systems at NASA if the organization didn’t use grid. Mehrotra, senior scientist and lead on the grid integrated project team at NASA Advanced Supercomputing Division, part of NASA’s Ames Research Center in Moffett Field, Calif., says grid computing gives his team a cost-effective means to store and maintain volumes of data used in projects examining the Earth, solar system and universe.
Grid computing pools processing cycles from multiple computers to maximize capacity, memory, power and other resources distributed across multiple systems. The concept of a grid describes a framework in which heterogeneous and distributed computational, networking, memory and storage resources can be linked to serve the needs of particular user applications, according to Nemertes Research.
Mehrotra says NASA started on its own with grid in 1996 and in 1997 began working with the Globus Alliance and its Globus Toolkit, an open source implementation for building grid-enabled applications. Grid proponents say the technology lets IT squeeze more out of their servers, storage and systems.
NASA initially deployed grid computing to share resources between locations, but now plans to use the technology to support critical projects and space missions. “Our original goal for investigating grid computing was to make more efficient use of NASA’s computational resources and to allow scientists and engineers easier access to these resources,” Mehrotra says. “[Now] we are investigating the use of these technologies for mission operations; both for mission and payload control and management.”
Gaining momentum
Once primarily a tool for leading-edge research engineers, grid is becoming more mainstream for enterprise IT departments in aerospace companies in particular. Bob Parker, industry analyst at AMR Research, says aerospace firms use grid more than other sectors because of the nature of their work. “Designing an aircraft carrier requires a lot of computing cycles, for example, so you will see these companies making a concerted effort with grid computing around a certain task,” Parker says.
According to Insight Research, total worldwide grid spending will increase from $250 million in 2003 to approximately $4.9 billion in 2008. It forecasts enterprise grid deployment to shift from early adoption to more widespread pilots and production launches this year.
A grid computing system manages two supercomputers at NASA’s Metacenter. The Metacenter is an exploratory project that aims to make NASA supercomputers more readily available to researchers, thus providing quicker turn-around for batch jobs, a larger range of available resources for computation and better distribution of the computational workload across multiple supercomputers. Grid computing lets jobs be migrated between the systems based on load and by using a scheduling system called Portable Batch Systems.
“We are a research and development center as well as a provider of services to NASA, so we tend to get involved in promising technologies early or create them ourselves,” Mehrotra says. “Before grid computing, NASA’s resources were isolated from each other or clustered in small groups. This makes it difficult for users to use resources at multiple sites.”
The team at aerospace manufacturer Boeing’s Phantom Works research and development division also uses grid computing to pool and distribute resources. John Hurley, a senior manager at Boeing Phantom Works in Seattle, is responsible for distributed systems integration and managing the group that focuses on grid computing. His group uses grid computing between Boeing sites in Puget Sound, Wash., and St. Louis. The primary reason Boeing started working with grid computing five years ago was to cut costs, but the technology also helped Boeing balance computing resources between the two sites.
“Grid computing takes advantage of different resources at different sites and doesn’t have any ownership or maintenance costs associated with it at the sites,” Hurley says. “Very few industries rely on computing as much as we do. We have large problems that need to be dealt with in real time.”
Boeing plans to expand the grid deployment to exploit the shared resources to load balance jobs among more locations and diminish bandwidth requirements on the network.
One obstacle Boeing faced in grid implementation was cultural more than technical. “It was a challenge to deal with a user community that was apprehensive about adopting a new way of computing, not just a new application,” Hurley says. The organization overcame this problem by offering extensive training on the new system.
Hurley couldn’t discuss the financial details of Boeing’s grid investment. However, he says the deployment removes the costs associated with maintaining separate resources and prevents one site from monopolizing resources, which enables sharing resources among many sites. “We had to get people looking at the bigger picture, how can we integrate everything together to work for the company, rather than how do my resources support my job and my users,” he says.
NASA’s Mehrotra encountered similar obstacles. “One change was the amount of coordination needed between NASA organizations. This coordination is needed to manage the common grid deployed by the organizations to ensure that compatible versions of grid software are installed, problems reported by users are resolved and so on,” Mehrotra says.
Aside from cultural issues, IT departments looking to adopt grid computing should be aware that the technology could become a “big ticket item” for companies not prepared to roll out grid applications, according to Hurley and Mehrotra. While grid is based mostly on deploying software applications that share resources on multiple systems, consistency and standardization is necessary across the infrastructure and other platforms supporting grid applications. Grid deployment also requires application code that supports parallel processing.
| |||||||||
Challenges aside, the two organizations will continue to work toward their goals for grid computing. “Our ultimate goal is that our users don’t even notice that grid computing is there. They can simply perform their scientific simulations, access their data and accomplish their work using large and diverse pools of resources and services without knowing the complexities involved,” Mehrotra says.
As for Boeing, the aerospace leader expects to see more commercial companies adopt grid computing in the near future.
“Grid has suffered with a reputation as a play tool for academics, but businesses are starting to recognize it’s a viable product and technology,” Hurley says.
AMR’s Parker also expects to see more grid computing deployments, but not in its purest definition. Even today, aerospace companies are pooling inexpensive resources for specific tasks rather than rolling out an on-demand grid across their entire infrastructures. “But it is grid in that it’s taking computing power off the desktop and centralizing it for economy, efficiency and mobility,” he says.
