“Making serial software-programs where a series of operations happen in strict sequential order— exploit multicore processors is a pressing problem. A group at North Carolina State University has developed a technique that gives these serial programs a greater degree of parallelism automatically, yielding performance improvements of some 20 percent in real-world programs.”
Here is some more Tech News from around the web:
- Microsoft: Office 15—not 2010—to be fully OOXML compliant @ Ars Technica
- System Update Re-Enables Hard Drives on Banned Xbox 360 Consoles @ NGOHQ
- Intel’s Atom D525 to offer 1.8GHz with no bump in consumption? @ Engadget
- Windows 7 SP1 Leaks to the Internet in Early Form @ DailyTech
Source: Ars Technica
If you want to design a killer app you won’t go wrong is designing ways to easily implement a way for programs to leverage better performance out of multicore systems. We have had singe core systems since the beginning so that is predictably where programmers have focused their efforts. There have always been specialized applications for servers that have more than one physical processor but until very recently there has not been the amount of multicore processors out there and as is usual in the computer world the amount of cores is increasing very quickly. Enter some researchers from N. Carolina State University who are working on new memory management techniques that offer an easy 20% jump in performance by offloading the work of deallocating memory to an idle core giving the ones under load more spare cycles. Ars Technica also points out how this performance tweak is also going to enhance security.