Adaptive Computing is upgrading its Moab management software to address the rapidly growing number of computing nodes and processor cores-and the increasing use of such accelerators as graphics chips-in the high-performance computing space.
Adaptive Computing is rolling out Moab 6.0 and its companion, Moab Viewpoint 2.0, both of which are designed to improve and streamline the command communications and reporting processes that in the past have made it difficult for HPC systems to scale beyond a few thousand nodes.
"We're address the extreme scalability and extreme usability of HPC systems," Peter ffolkes, vice president of marketing at Adaptive Computing, said in an interview with eWEEK.
Adaptive unveiled the enhanced offerings Nov. 15 at the Supercomputer 2010 show in New Orleans.
HPC cluster systems are growing in breadth and complexity, adding nodes and cores at a rapid rate and forcing management software tools to keep up, ffolkes said. HPC clusters are scaling from tens of thousands of cores to hundreds of thousands, making them "quite complex to manage," he said.
Moab 6.0 offers a 100-fold improvement in the speed and availability of state-and-command communication management, ffolkes said. It leverages multi-threading technology and a database of real-time job and node status data, and enhancements include support fro the most commonly used Moab commands and grids that deplye multiple instances of Moab.
It also offers support for hybrid clusters that use both CPUs and GPUs (graphics processing units) in a general computing environment. Using GPUs as co-processors helps the systems boost performance and improve parallel computing capabilities while helping to hold down power and space costs, and support for such hybrid systems in a management solution is key. The latest list of the top 500 fastest supercomputers illustrates the rapid rise in the use of such accelerators.
For example, 17 systems in the Top500 list use GPUs as accelerators from such vendors as Nvidia and Advanced Micro Devices, which is building out the graphics business that it acquired through its purchase of ATI in 2006.
The Tianhe-1A system at the National Supercomputer Center in China, which with a performance level of 2.57 petaflops (or quadrillions of floating point calculations per second) is now the fastest supercomputer in the world, uses Nvidia GPUs as accelerators. The third-fastest supercomputer, the Nebulae, also in China, and the Tsubame 2.0 at the Tokyo Institute of Technology in Japan, at number four, also use Nvidia graphics chips.
Adaptive Computing's ffolkes said that like the increase in cores and nodes in HPC clusters, the growing use of accelerators also increases the complexity of the systems and challenge to management tools to handle that complexity.
In addition, Moab 6.0 offers new on-demand dynamic provisioning and management capabilities for both physical and virtual environments, tools to improve utilization and energy efficiency of the systems, and improved administration and reporting capabilities.
Moab Viewpoint 2.0 enables users to offer HPC capabilities in an enterprise cloud computing environment, including physical and virutal node management for extremes-scale environments, dynamic management of virtual machines and physical nodes for HPC and enteprise clouds, support for hundreds of physical nodes and thousands of virtual machines, and service-based administration and reporting capabilities.