Taking the idea of peer-to-peer resource sharing to the level of secure, enterprise-quality grid computing, Entropia will soon release its DC (desktop computer) Grid 5.0 software suite, which delivers tools to help companies harvest compute cycles from a grid of desktop PCs, according to Andrew Chien, the CTO of Entropia, in San Diego.
Speaking here at the Global Grid Forum, Chien said DC Grid 5.0 will allow companies to run a wide range of applications using the idle compute cycles of their PC fleets, without having to integrate the applications into the full operating environment of the PCs.
The result will be a grid of PCs that could save many companies the expense of purchasing clustered servers or other, similar supercomputing devices.
DC Grid 5.0's launch is scheduled to happen before mid-2002, said Chien.
"We've done a large number of deployments in which we've measured the gathered cycle efficiency as over 90 percent," said Chien of DC Grid 5.0. "In some cases you can do simple arithmetic and see that there is a huge amount of computing power available with desktop grids."
Running jobs using CD Grid 5.0 does not interfere with the primary PC user experience, and applications are run at a binary level, in a "binary sandbox," using the same type of open job scheduling used by large, shared server clusters and scientific computing grids, said Chien.
"The binary sandbox ensures a set of principals for isolation between the application and the underlying environment," he said. "It also provides pretty strong security guarantees for protecting the application from tampering in the external environment."
DC Grid 5.0 is "language-neutral," said Chien, and can support codes such as C, C++, Fortran, Java C#, and others.
A range of applications can also be handled within a PC grid network using DC Grid 5.0, including scientific applications, Monte Carlo simulations, and many data-intensive business applications, said Chien.
"Not all applications scale [over distributed grid networks], but many do," said Chien. "And the floating point performance of x86 processors [inside the PCs] is quite impressive these days."
"One of the reasons we are pursuing this enterprise model [with PC grids] is when you go into a enterprise, it's relatively easier [than a distributed set of servers]," said Chien. "There is often a command structure that supports some policy that says these machines are available."