Ryan Waite, provides overview of Microsoft Technical Computing and HPC Server 2008R2. The sheer size of generated data is a key driver in today’s computational ecosphere. By the end of 2010, over 1.2 zetabyte of data will have been generated. Microsoft has been working closely with Intel to integrate their Parallel Studio product into Visual Studio 2010. R2 is Microsoft’s third release of HPC product built on Windows, scalable to thousands of nodes.
Matt Blythe, offers an introduction to Dual-Boot “Hybrid” Clusters that allow Windows HPC and Linux HPC to exist on same hardware. A key advantage of this approach is simplified integration into existing Linux-centric IT environments. It also enables OEM partners to continue providing platform-centric solutions without expending additional effort to port codes across platforms.
Hybrid Windows HPC & Linux Clusters:
Eric Lantz, outlines some tools that can be used to optimize HPC Cluster performance. Three usage patterns are considered: Data Parallel, Message Passing (MPI), and SOA (Service-Oriented Architecture). The biggest gains can be made in Networking optimization. R2 features Network Direct, a new networking interface that runs on RDMA-capable hardware over Infiniband, 10 Gigabit Ethernet, as well as other Layer 2 architectures.
Performance Tools for Windows HPC Server 2008R2:
Integration of Excel 2010 and HPC Server 2008 R2 maintains the Excel front end but offloads the calculation workload to cluster. The result is a reduction of compute time by a factor of ten on a modest four-node cluster.
Excel services for Windows HPC Server 2008R2:
Rae Wang, enumerates the diagnostics and reporting features available in HPC Server 2008 R2. Microsoft designed comprehensive diagnostics to assist with granular troubleshooting, empowering administrators, ISVs, and IHVs to address failures. Rae gives a walk-through on the UI and discusses Extensibility features implemented in Diagnostic console and Reporting features. HPC exposes reporting data, making it available for data warehousing, manipulation with Excel, or other authoring tools. For this purpose, the available data interfaces are Powershell and database views.
Diagnostics and reporting in Windows HPC Server 2008 R2:
Greg Burgess, describes HPC job scheduler. Job lifecycle is described according to three phases, Admission (jobs enter the scheduler), Allocation (scheduler makes decisions about job placement), and Activation (scheduler starts and controls the job).
HPC Server 200R2 Job Scheduler:
Kathy Palmer, presents the management and monitoring enhancements released with R2. The notion of personal supercomputing is a core operational paradigm that’s rapidly evolving in the numerical computing ecosphere. Various visualization aids, such as heat maps, provide administrators with hierarchical and aggregate views that simplify utilization, health management, and troubleshooting.
Windows HPC Server 2008R2 Monitoring and Management Part 1:
Part 2: explores cluster management and demonstrates how major administrative tasks are performed from a single integrated console: Configuration & Deployment, Monitoring, Node Management, Job Management, Reporting, and Diagnostics.