Altair Unveils PBS Professional 11.0, Optimized PetaFLOPS Scalability, GPU and Topology Scheduling
Scheduling Jobs onto NVIDIA Tesla GPU
Computing Processors using PBS Professional
Major features and benefits of PBS Professional 11.0 include:
- Faster Scheduling: Scheduler speed is up to three times faster, down to less than 30 seconds per cycle for typical HPC workloads (on up to 80,000 cores).
- Faster Submission: Job submission speed is seven times faster.
- Faster Cold Start: Cold start is 45 times faster, reduced to less than one minute for 100,000 jobs.
- Re-architected Cray Support: PBS Professional "vnode" support for Cray systems, allowing individual MPI task selection, task placement, support for node exclusivity, and the framework for topology-aware scheduling.
Additionally, PBS Professional meets a growing demand for:
- Cloud Computing: HPC Software-as-a-Service (SaaS) Clouds are being powered by the full PBS Works suite, including Altair's own HyperWorks On-demand.
- Topology-aware Scheduling: With the addition of Cray support, PBS Professional now optimizes task placement for all HPC network topologies – such as Infiniband, SGI, Cray, IBM and GigE – improving application performance and reducing network contention.
- GPU Scheduling: Leveraging PBS Professional's extensible architecture allows for graphic processing units (GPUs) scheduling on all platforms.
“The future of GPU technology for HPC as a general-purpose computing unit is becoming increasingly important as the industry moves toward GPU-based clusters for running scientific and engineering applications,” said PBS Works Chief Technical Officer Bill Nitzberg, Ph.D. “The ability for a workload management tool, like PBS Professional, to support GPU configuration and scheduling is essential to enable this type of computing. With PBS Professional, customers are able to use a CPU and GPU together in a heterogeneous computing model, greatly increasing application speed and lowering overall power requirements.”
In October 2010, PBS Works announced the selection of PBS Professional by NEC Corporation (NEC) as the workload management tool for TSUBAME 2.0, one of the world’s fastest supercomputer systems. TSUBAME 2.0 is extremely GPU-rich and incorporates many of the latest green and cloud technologies. PBS Professional will administer up to 100,000 jobs at a time for the supercomputer system, which contains nearly 20,000 processing cores and 4,200 NVIDIA Fermi GPUs to generate 2.4 petaFLOPS (a thousand trillion floating point operations per second) of computing power.
“Another key trend for HPC is SaaS through cloud computing,” said Nitzberg. “By running software applications through the cloud, companies can free up computing power of HPC systems, enabling increased application speed and lower power usage. Altair is currently developing a new cloud-computing infrastructure that will allow customers to have access to the PBS Works suite of software as well as Altair server infrastructure using Altair’s unique, patented units-based licensing model.”
PBS Works also offers other HPC solutions within its suite: PBS Catalyst and PBS Analytics. PBS Catalyst simplifies the submission and management of jobs running on PBS Professional compute nodes. Since there are no complicated commands to learn or scripts to write and maintain, PBS Catalyst increases productivity for users by allowing more time for value-added research and analysis. To best meet demanding and varied requirements, PBS Catalyst comes as two modules: a Web-based version, allowing users to easily submit work to remote PBS Professional compute nodes, and a powerful desktop client version, enabling users to submit jobs to local and remote systems.
PBS Analytics is a data analysis and visualization solution that provides PBS Works administrators with advanced job and license analyses to support data-driven planning and decision-making. These Web-based portals support multiple simultaneous views of historical usage data.
PBS Analytics includes two modules: License Analytics, enabling customers to maximize software license use by allowing charge-backs to cost centers, greater accuracy in capacity growth planning, and reduction of spending via sharing of licenses. Workload Analytics provides HPC administrators with a suite of simple, customizable tools for analyzing and viewing complicated job data, providing visibility into how well expensive HPC resources are being utilized.