New Approaches to Parallelism

Share it on Twitter  
Share it on Facebook  
Share it on Linked in  

Intel is taking additional steps to avoid what could be a serious bump in the road to convert the data center to multicore technology.


The fact that software developers are having trouble cranking out new versions of their systems to take advantage of multicore's parallel processing capabilities hasn't gone unnoticed. Intel, in fact, was one of the first to sound the alarm that the development community needs to embrace new programming techniques like multi-threading if it hopes to push the envelope that multicore provides.


Now, the company is going one better with a new technology said to deliver the power of parallel processing with the scalability of linear technology. The company has joined forces with Rogue Wave, developer of the C++-based Hydra system that brings "service parallelism" to the new Xeon Server (Dunnington) platform. The companies said the Hydra system allowed a 24-core server to exceed the expected three-fold increase in performance compared to an eight-core device. What's more, the system works without modifying the application software -- so no more multi-threading.


Service parallelism is described here as a cross between SOA and parallelism. Rather than running application loops or functions in parallel, service parallelism runs multiple instances of an actual service. Besides not having to recode software, the technique is also said to improve application maintenance.


Intel hasn't given up on multi-threading entirely, however. The company recently introduced the Parallel Studio module for Microsoft Visual Studio environments. The system scans applications to determine which code needs to be updated, and provides things like a library and debugging tools to help smooth out the process.


Still, multi-threading may become a tougher sell in the near future, not just to developers but to end users as well. The driving force behind parallel programming has always been to allow data centers to leverage all of the cores they've invested in. If a serial program ties up one core on a multicore device, then the remaining cores generally sit idle until they receive instructions of their own. But VMware, for one, is looking to change that. In its drive to make the OS obsolete before Microsoft and Sun do the reverse, the company is touting its new Virtual Datacenter OS (VDC OS). Among its many attributes, VDC OS can allocate computing resources on demand, so even if serial applications are using one core, the remainder will still be available for other workloads.


In the end, though, it's likely that most enterprises will be eager to embrace parallel simply because of the sheer increase in processing power it offers. This video shows a demonstration that the Mythbuster guys, Adam Savage and Jamie Hyneman, gave to a recent Nvidia conference. If you can imagine the same thing happening on your network, you'll know why the powers-that-be are so jazzed about parallel.