A startup has come out with software that promises to overcome a major grid computing headache: namely, converting applications so they can be spread across a grid or cluster with standard machines, as opposed to exotic, high-end computers.
Aspeed Software Corp.s Accellerant 3.1, a software development kit that the New York company announced on Tuesday, includes APIs and services that can be embedded into existing code or used to wrap binary instances of the application that needs to be parallelized for multi-CPU, cluster and grid configurations.
Kurt Ziegler, executive vice president of development, said the technology is designed to tackle the thorny issue of architecture within applications theyre trying to get onto grid or other distributed environments.
While such companies focus on database performance and response times, the real key is getting the applications to stop making so many database calls in the first place, he said.
“I quiz people,” he said. “I ask, What is the optimum number of database calls an application makes? They scratch their heads. Zero is the answer. Theyre all focused on response times and performance. If I had it in memory and didnt have to make a database call, ideally, if you talk to [database administrators], theyll tell you very quickly: Their whole focus is optimizing response times.”
If applications are designed to initiate 100 database calls, theres not much anybody can do at the grid or database level to change the profile of the application from a performance standpoint, Ziegler said. “If youre going to distribute, you want to be able to have the application be enabled to take advantage of distribution: across grid, across a cluster, across multiple core CPUs.”
In focusing on the applications themselves, Accelerant 3.1 provides a framework to distribute existing single-thread and multithread applications to distributed environments built on Linux and .Net and also fits into SOA (service-oriented architecture) environments. It depends on patent-pending algorithms to provide adaptive workload distribution and redistribution for optimum run times, data movement capability, serialization and unserialization, and compression services.
Carl Claunch, an analyst at Gartner Inc., said he likes Aspeeds solution to this perennial problem, particularly as it will offer a more affordable option for sectors that need it badly, such as the energy industry.
“One of the big challenges in grid becoming more [available to a general audience] is to be able to convert more applications so you can more quickly achieve more accuracy or take on bigger problems with spreading it across more machines,” said Claunch, in Los Altos, Calif. “Its been decades of work to find [answers to] parallel problems. Thats pretty well-developed today, but lots of areas are still really frustrating.
“The energy sector, for example … to exploit new reservoirs of oil and gas, they have to continue to use expensive, exotic, high-end supercomputers. Aspeed is promising tools to help those people find a way to convert applications so theyll get a reasonable return by spreading them across a grid or cluster with more standard-volume machines.”
Typically, Claunch said, such capabilities are built in-house. Thats OK for Wall Street, where firms typically have plenty of mathematicians on staff, but other sectors need more help. As it now stands, many must rely on third-party companies to figure out how to perform the calculations for applications that will enable them to perform well on grid.
Accellerant requires no dedicated management server, since its application-based. It supports multiple platforms running Linux Red Hat, Linux SuSE, Windows NT and XP, and Unix Solaris. Its directly available now from Aspeed.