Current OS research in HPC varies in approach, ranging from custom lightweight solutions such as Sandia's Catamount OS (ref) to scalable Linux variants, like ZeptoOS (ref) and Cray's Ranier (ref). Current experience suggests there is no on-size-fits-all OS solution, with each OS carrying advantages for certain targeted scientific applications. Furthermore, it is not clear which OS is better suited for current terascale or future petascale class systems. In order for these systems to run “out-of-box” several challenges in system software and application runtime environments have to be addressed. Efficiently exploiting tens-to-hundreds of thousands of processor cores using tens-to-hundreds of interdependent computational tasks requires appropriate scalability, manageability, and ease-of-use at the system software and application runtime environment level. Furthermore, given the expected system upgrade interval, the design of this system software and runtime environments demand an approach that avoids excessive porting.

• Kurt Ferreira
• Zheng Cui

• Main development is at our separate Project Home Page
• The Kitten kernel from Sandia is forming the basis for our work on virtualization for HPC systems
• A description of the development setup at UNM is available elsewhere on this wiki.