Mass-Market Multicore

Computerworld - For vendors and end users alike, the ascent of multicore processors is as beneficial as it is inevitable. Vendors (of chips, hardware, software, networking gear -- the works) will get to roll out products with a high buzz factor, while users will revel in the seemingly ceaseless Moore's Law march: ever-cheaper, ever-faster, ever-better computing.

Stuck in the middle are IT professionals, who must make it all work.

The writing is on the wall: Multicore chips will soon dominate the technology landscape. The good news is that for now, developing software for multicore processors differs little from developing for the multiprocessing systems with multiple chips in common use. Boosting performance by adding chips to achieve symmetric multiprocessing is an old IT standby, particularly in database servers, heavy number-crunching and other data-intensive applications.

In addition, as multicore evolves, look for vendors to offer tool sets that help you optimize your code for multicore, as well as prewritten code modules that remove the burden entirely. Vendors, like enterprise developers, will also hone their ability to write applications that take advantage of multicore's inherent advantages.

While IT professionals experienced with servers won't see a major difference between multiprocessor and multicore work, those whose experience is limited to the desktop - where multiprocessing is an alien concept -- will have to adapt. "As dual-core gets out, it'll be the first time multiprocessing ever hits the mass market," says Alan Zeichick, a multicore development expert and principal analyst at Camden Associates in San Bruno, Calif. "People who have always developed for desktops ... will need to learn to write threaded apps."

Enterprise developers may also find new challenges in partitioning applications. "C and C++ are sequential languages with no concept of running in parallel," says Sven Brehmer, CEO of PolyCore Software Inc., a Foster City, Calif.-based vendor of multicore development tools. To partition an application written in these languages, you must break the application apart, then run some portions of it on one processor and other portions on another. "When you take these parts and move them, you need to make sure you don't destroy variables or the way data flows from the source to the destination," Brehmer says.