High-performance computing

Introduction to High-performance Computing (HPC)

High-performance computing (HPC) leverages supercomputers and computer clusters to tackle advanced computational problems. HPC is distinct from high-throughput and many-task computing, focusing instead on executing complex calculations at high speeds.

In this document you will read:

Overview of HPC

HPC merges several disciplines including systems administration (encompassing network and security), parallel programming, digital electronics, computer architecture, system software, programming languages, algorithms, and computational techniques.

Tools and Technologies in HPC

HPC technologies are essential for developing and implementing high-performance computing systems.

Recently, the focus within HPC has shifted from traditional supercomputers to computing clusters and grids, facilitated by advancements in networking technologies such as the use of collapsed network backbones for easier troubleshooting and upgrades.

Scientific and Business Applications of HPC

Historically associated with scientific research, HPC now also encompasses high-performance technical computing (HPTC), which includes engineering applications like computational fluid dynamics and building virtual prototypes. Moreover, HPC applications extend to business operations such as data warehousing, transaction processing, and line of business (LOB) applications.

Evolution and Context of HPC

The term HPC emerged post the advent of supercomputing, sometimes used interchangeably with supercomputing, though it specifically refers to a broader range of powerful computing solutions.

Challenges in Scaling HPC Applications

Many current applications are not originally designed for HPC but have been adapted, often leading to issues when scaled to more powerful processors or larger systems.

Practical Applications of HPC Technologies

Simulation of car crashes for automotive design.
Molecular interaction simulations for new drug development.
Analysis of airflow over vehicles and aircraft.

TOP500

Overview of TOP500

The TOP500 list ranks the world's 500 fastest high-performance computers based on the High Performance LINPACK (HPL) benchmark. This ranking includes only those systems whose owners have submitted an HPL score. Not all high-performance computers appear on the list, either due to ineligibility to run the HPL benchmark or because owners choose not to disclose their system's capabilities, often for security reasons.

Criticism of the LINPACK Benchmark

The reliance on the LINPACK benchmark has sparked controversy, as it cannot comprehensively evaluate all aspects of a high-performance computer's capabilities. This limitation has prompted discussions on the need for more diverse benchmarks that can provide a fuller picture of a system's performance across different computing tasks.

Applications of High-Performance Computers in the TOP500

High-performance computers listed in the TOP500 are utilized in a variety of critical research and development areas. For instance, scientists use these supercomputers for simulating galaxy creation, fusion energy, and global warming. These systems are also crucial for enhancing the accuracy of both short-term and long-term weather forecasts.

One notable example is the IBM Roadrunner, which in 2008 was the world's tenth most powerful supercomputer. Located at the United States Department of Energy's Los Alamos National Laboratory, it was used to simulate the performance, safety, and reliability of nuclear weapons, ensuring their operational functionality.

High Performance Computing in the Cloud

Introduction to Cloud-based High Performance Computing

High performance computing (HPC) in the cloud integrates the power of traditional HPC systems with the flexibility of cloud environments. This combination allows for scalable computing resources that can be adjusted according to the demands of specific tasks and projects.

Evolution of HPC Benchmarks

As part of efforts to address the limitations of the LINPACK benchmark, which is notably used in the TOP500 ranking of supercomputers, the U.S. government commissioned Jack Dongarra from the University of Tennessee to develop a more comprehensive set of tests. This led to the creation of the HPC Challenge benchmark suite, which includes LINPACK among other tests. Although this suite offers a broader evaluation of HPC capabilities, it has struggled to gain the same level of recognition as the single-number LINPACK test used in TOP500 rankings. The TOP500 list is updated biannually, reflecting new data from the ISC European Supercomputing Conference in June and the U.S. Supercomputing Conference in November.

Grid Computing and HPC

The concept of grid computing, which serves as a precursor to modern cloud-based HPC, has borrowed many principles from traditional high performance computing. This approach has laid the groundwork for the current implementations of HPC within cloud infrastructures, emphasizing enhanced scalability, flexibility, and resource management.

Key Benefits and Challenges

Utilizing cloud environments for high performance computing presents several advantages, including the ability to scale resources rapidly and cost-effectively. However, the integration of HPC into cloud platforms also brings challenges, particularly in terms of achieving the high levels of performance and efficiency typically associated with dedicated supercomputing hardware.

Stories

High-performance computing