HPC for the Rest of Us - Part 1

Understanding HPC High Performance Computing (HPC) is about computing that pushes the limits of available performance. In meeting our ever-increasing needs for new quantitative precision and complex systems modeling, HPC lets us tackle much bigger problems with the same (or less) amount of time than we use today for much smaller problems. From meteorological forecasting and automotive crash simulations to human genome mapping and gene sequencing, HPC delivers the ability to do bigger, better, and more exciting science. Unlike supercomputing, HPC achieves this with a broad range of systems, not just "big iron" computers. HPC is based on parallel computing, the primary goals of which are to cut turnaround time (i.e. more computations, less time), and accommodate job upsize (i.e. scale for more complex scenarios). Parallel processing, or parallelism, divides program instructions among multiple processors to run a program in less time. A key principle here is concurrency, or performing multiple tasks at the same time. This is achievable with either throughput computing or turnaround computing. In throughput computing, a single HPC infrastructure addresses multiple individual jobs from multiple parties to achieve the desired result. Turnaround computing involves processing a single job before addressing the next. In hardware terms, parallel computing is available in three forms: on chip: e.g., on Itanium processor-based platforms and Explicitly Parallel Instruction Computing (EPIC) multiprocessor: symmetric multi-processor (SMP) machines multi-computer: clusters The demand for HPC continues to surge in both enterprise IT and technical computing environments as we seek to understand and control increasingly complex processes. Designed for highly efficient parallelism, the EPIC model enables processor performance levels to scale by improving parallel instruction throughput. With EPIC, the processor is relieved from having to identify parallel-processing opportunities, freeing both hardware and software to do what they do best. In the last few years, parallel computing has entered the mainstream with the wide adoption of SMP shared memory machines, where multiple processors share a common operating system and memory. Clusters and Grids One of the most important computing trends to emerge in the new millennium is the movement away from the large servers and monolithic supercomputers of the 70s to modern clustered solutions. Clusters acting as a single highly available system make HPC more affordable. Rising performance, standardization, and plunging costs are just some of the factors that are driving the demand for HPC beyond technical scientific research and into mainstream business applications, such as financial economic modeling and large financial risk management and business intelligence. (Thus, the difference of HPC and HPTC (High Performance Technical Computing), where the earlier refers to more business orientated usage/applications). This trend has accelerated with companies like Intel increasing focus on delivering commercial, off-the-shelf (COTS) technology solutions, and investing in various initiatives to bring together software developers and service providers to collaborate on deploying and supporting these demanding systems. Today, turnkey Intel-powered COTS cluster systems are available from several leading manufacturers and specialty solution providers, complete with polished releases of commercial software and comprehensive product support. Businesses are racing to deploy them based on a nearly irresistible line-up of features and benefits. HPC has found a welcome home in such industries as manufacturing, finance, energy, life sciences and digital media in addition to the public sector in government, education and national labs. COTS systems continue to drive the growth of cluster computing, scaling huge amounts of compute capacity at more competitive price points. Proprietary systems too have evolved towards open standards, custom-based stacks. Click here to see Intel's Supercomputer roots. By combining the power of desktops, clusters and large SMP systems, HPC is approaching the era of the Grid. This extends the notion of open, standards-based middleware to enable coordination among connected computing platforms - potentially anywhere in the world. By enlisting clients, departmental servers, large SMP systems and large data center servers, grids can provide affordable resources at unprecedented size and economies of scale. Most Grid projects today are aiming to create and share HPTC across distributed networks for large compute power needs). The Top 500 Supercomputer list* is a good indicator of how rapidly HPC has evolved. HPC typically involves clusters with at least one TFlops (Tera Floating-point Operations Per Second) of computing power. New data points now reflect supercomputer diversification, and performance differences between low and high-end models. Over the past year, Intel-based systems have surged from three to 119 entries in the Top 500 list. Of this, the Asia-Pacific region accounts for at least 15 Intel-based clusters. Rising performance, plunging costs and surging demand have combined to launch HPC rapidly beyond scientific research labs and into mainstream businesses. Many business problems may now be revisited using novel computing, simulation or optimization techniques, which will require powerful computers. Central to a global community of equipment manufacturers, software developers, system integrators and service providers, HPC installations have achieved superior performance, scalability, flexibility and upgradeability - at a fraction of the cost of proprietary RISC-based HPC clusters, mini-computers and supercomputers. The second part of this article will examine HPC applications to vertical markets, and the role of software. * Updated twice a year, the Top 500 table shows the 500 most powerful commercially available computer systems, and lists the sites that have these computer systems installed (http://www.top500.org/). Side Bar: HPC's Aims Minimize turnaround time to complete specific computing problems Maximize the problem size in a given amount of time, and solve it Tackle complex and large-scale problems that would otherwise be impossible Maximize price-performance (best Flops per dollar) Minimize costs, and the impact of obsolescence Extend to solve medium-sized problems (and budgets) Sidebar: Common HPC acronyms FLOPS: FLoating-point Operations Per Second. A measure of the capability of a machine or cluster to deal with scientific calculations. Parallel Computing: Calculating parts of the program at the same time to shorten the overall amount of time needed to solve the problem. MPP: Massively Parallel Processing. Linking together many (sometimes as much as 9000 processors) in a single machine through a specialized interconnect scheme. SMP: Symmetric Multi-Processing. A configuration of processors in a single computer that allows all processors the same access to system resources such as memory or I/O. In an SMP system, each processor is equally capable and responsible for managing the flow of work through the system. Cluster: A collection of connected independent computers that work in unison to solve a problem. SIMD: Single Instruction, Multiple Data. This is parallel computing architecture which performs one operation on multiple sets of data. In this instance, one processor is used for the control logic, while the remaining processors are used as slaves. MIMD: Multiple Instruction, Multiple Data. In this case, every processor is controlled independently, with each processing its own set of instructions simultaneously. High Availability: A form of clustering to give non-stop computing in the event of partial hardware failures. Load Balancing: A form of clustering that distributes an incoming workload across a series of computers. Usually in reference to incoming TCP/IP traffic to a web site. Sidebar: Further Reading General background: http://developer.intel.com/design/servers/ipmi/ http://www.intel.com/go/hpc/ http://www.beowulf.org http://www.microsoft.com/windows2000/hpc/ http://clusters.top500.org/ http://www.tc.cornell.edu/AC3/ (Advanced Cluster Computing Consortium) Vijay Keshav is the Industry Solutions Manager for the Asia Pacific region at Intel® looking after their High Performance Computing and Life Sciences businesses. His key role is to bring together the various Strategies and Solutions of Intel® and its ecosystem, towards offering customers a compelling value proposition of deploying such solutions on Intel® Architecture. An Electrical engineer with a Masters Degree in Management, he has worked in senior Sales, Marketing & Business Development roles with leading IT companies across the Asia Pacific region before joining Intel® Asia Electronics Inc in 2000.

HPC for the Rest of Us - Part 1

Related

Forget the Pixel 8a. This $399 Samsung phone is a force to be reckoned with

I fly 10 times a year. These 5 tech gadgets are lifesavers

This 11-in-1 docking station freed up my cluttered workspace, and it's 40% off