SCTuner: An Auto-tuner Addressing Dynamic I/O Needs on Supercomputer I/O Sub-systems.
Abstract not provided.
SEARCH · Engineering Papers
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
Abstract not provided.
Explore the source record for details and available documents.
In the span of four decades, quantum computation has evolved from an intellectual curiosity to a potentially realizable technology. Today, small-scale demonstrations have become possible for quantum algorithmic primitives on hundreds of physical qubits and proof-of-principle error-correction on a single logical qubit. Nevertheless, despite significant progress and excitement, the path toward a full-stack scalable technology is largely unknown. There are significant outstanding quantum hardware, fabrication, software architecture, and algorithmic challenges that are either unresolved or overlooked. These issues could seriously undermine the arrival of utility-scale quantum computers for the foreseeable future. Here, we provide a comprehensive review of these scaling challenges. We show how the road to scaling could be paved by adopting existing semiconductor technology to build much higher-quality qubits, employing system engineering approaches, and performing distributed quantum computation within heterogeneous high-performance computing infrastructures. These opportunities for research and development could unlock certain promising applications, in particular, efficient quantum simulation/learning of quantum data generated by natural or engineered quantum systems. To estimate the true cost of such promises, we provide a detailed resource and sensitivity analysis for classically hard quantum chemistry calculations on surface-code error-corrected quantum computers given current, target, and desired hardware specifications based on superconducting qubits, accounting for a realistic distribution of errors. Furthermore, we argue that, to tackle industry-scale classical optimization and machine learning problems in a cost-effective manner, heterogeneous quantum-probabilistic computing with custom-designed accelerators should be considered as a complementary path toward scalability.
Explore the source record for details and available documents.
High Performance Computing (HPC) is an important method for scientific discovery via large-scale simulation, data analysis, or artificial intelligence. Leadership-class supercomputers are expensive, but essential to run large HPC applications. The Petascale era of supercomputers began in 2008, with the first machines achieving performance in excess of one petaflops, and with the advent of new supercomputers in 2021 (e.g., Aurora, Frontier), the Exascale era will soon begin. However, the high theoretical computing capability (i.e., peak FLOPS) of a machine is not the only meaningful target when designing a supercomputer, as the resources demand of applications varies. A deep understanding of the characterization of applications that run on a leadership supercomputer is one of the most important ways for planning its design, development and operation. In order to improve our understanding of HPC applications, user demands and resource usage characteristics, we perform correlative analysis of various logs for different subsystems of a leadership supercomputer. This analysis reveals surprising, sometimes counter-intuitive patterns, which, in some cases, conflicts with existing assumptions, and have important implications for future system designs as well as supercomputer operations. For example, our analysis shows that while the applications spend significant time on MPI, most applications spend very little time on file I/O. Combined analysis of hardware event logs and task failure logs show that the probability of a hardware FATAL event causing task failure is low. Combined analysis of control system logs and file I/O logs reveals that pure POSIX I/O is used more widely than higher level parallel I/O. Based on holistic insights of the application gained through combined and co-analysis of multiple logs from different perspectives and general intuition, we engineer features to "fingerprint" HPC applications. We use t-SNE (a machine learning technique for dimensionality reduction) to validate the explainability of our features and finally train machine learning models to identify HPC applications or group those with similar characteristic. To the best of our knowledge, this is the first work that combines logs on file I/O, computing, and inter-node communication for insightful analysis of HPC applications in production.
This document is the final scientific report of the project DE-SC000120 (A scalable Development Environment for Peta-Scale Computing). The objective of this project is the extension of the Parallel Tools Platform (PTP) for applying it to peta-scale systems. PTP is an integrated development environment for parallel applications. It comprises code analysis, performance tuning, parallel debugging and system monitoring. The contribution of the Juelich Supercomputing Centre (JSC) aims to provide a scalable solution for system monitoring of supercomputers. This includes the development of a new communication protocol for exchanging status data between the target remote system and the client running PTP. The communication has to work for high latency. PTP needs to be implemented robustly and should hide the complexity of the supercomputer's architecture in order to provide a transparent access to various remote systems via a uniform user interface. This simplifies the porting of applications to different systems, because PTP functions as abstraction layer between parallel application developer and compute resources. The common requirement for all PTP components is that they have to interact with the remote supercomputer. E.g. applications are built remotely and performance tools are attached to job submissions and their output data resides on the remote system. Status data has to be collected by evaluating outputs of the remote job scheduler and the parallel debugger needs to control an application executed on the supercomputer. The challenge is to provide this functionality for peta-scale systems in real-time. The client server architecture of the established monitoring application LLview, developed by the JSC, can be applied to PTP's system monitoring. LLview provides a well-arranged overview of the supercomputer's current status. A set of statistics, a list of running and queued jobs as well as a node display mapping running jobs to their compute resources form the user display of LLview. These monitoring features have to be integrated into the development environment. Besides showing the current status PTP's monitoring also needs to allow for submitting and canceling user jobs. Monitoring peta-scale systems especially deals with presenting the large amount of status data in a useful manner. Users require to select arbitrary levels of detail. The monitoring views have to provide a quick overview of the system state, but also need to allow for zooming into specific parts of the system, into which the user is interested in. At present, the major batch systems running on supercomputers are PBS, TORQUE, ALPS and LoadLeveler, which have to be supported by both the monitoring and the job controlling component. Finally, PTP needs to be designed as generic as possible, so that it can be extended for future batch systems.
Supercomputing systems are used for a wide range of computationally demanding tasks in many fields of science and engineering. They play a key role in numerical simulation, in which mathematical models are computed in order to simulate the behavior of physical systems. Scientists and engineers that use supercomputers for numerical simulation often have their productivity limited by the need to manually organize and manage extremely large amounts of data that are often produced and consumed by the software programs run on these systems. Recognizing these limitations, Kitware Inc. (Clifton Park, NY) and SLAC National Accelerator Laboratory (Menlo Park, CA) are developing an advanced software platform that can reduce the cognitive overhead required by knowledge workers when using supercomputers for numerical simulation. Phase I of the project is complete and includes the development of new capabilities for organizing simulation project files, improvements to the user interface and overall usability, and deployment of a “middle tier” server to sit between user desktop machines and supercomputers to offload much of the data management workload. The project also developed prototype software for executing sequences of numerical simulations, and a prototype for migrating supercomputing software to cloud-based computing systems to provide a potential alternative to supercomputers with different logistical and price-to-performance tradeoffs.
NASA created the Numerical Aerodynamic Simulation (NAS) Program in 1987 to focus resources on solving critical problems in aeroscience and related disciplines by utilizing the power of the most advanced supercomputers available. The NAS Program provides scientists with the necessary computing power to solve today's most demanding computational fluid dynamics problems and serves as a pathfinder in integrating leading-edge supercomputing technologies, thus benefiting other supercomputer centers in Government and industry. This report contains selected scientific results from the 1991-92 NAS Operational Year, March 4, 1991 to March 3, 1992, which is the fifth year of operation. During this year, the scientific community was given access to a Cray-2 and a Cray Y-MP. The Cray-2, the first generation supercomputer, has four processors, 256 megawords of central memory, and a total sustained speed of 250 million floating point operations per second. The Cray Y-MP, the second generation supercomputer, has eight processors and a total sustained speed of one billion floating point operations per second. Additional memory was installed this year, doubling capacity from 128 to 256 megawords of solid-state storage-device memory. Because of its higher performance, the Cray Y-MP delivered approximately 77 percent of the total number of supercomputer hours used during this year.
In several cases, new visualization techniques have vastly increased the researcher's ability to analyze and comprehend data. Similarly, the role of networks in providing an efficient supercomputing environment have become more critical and continue to grow at a faster rate than the increase in the processing capabilities of supercomputers. A close relationship between scientific visualization and high-speed networks in providing an important link to support efficient supercomputing is identified. The two technologies are driven by the increasing complexities and volume of supercomputer data. The interaction of scientific visualization and high-speed networks in a Computational Fluid Dynamics simulation/visualization environment are given. Current capabilities supported by high speed networks, supercomputers, and high-performance graphics workstations at the Numerical Aerodynamic Simulation Facility (NAS) at NASA Ames Research Center are described. Applied research in providing a supercomputer visualization environment to support future computational requirements are summarized.
The NAS Parallel Benchmarks (NPB) are a suite of parallel computer performance benchmarks. They were originally developed at the NASA Ames Research Center in 1991 to assess high-end parallel supercomputers. Although they are no longer used as widely as they once were for comparing high-end system performance, they continue to be studied and analyzed a great deal in the high-performance computing community. The acronym 'NAS' originally stood for the Numerical Aeronautical Simulation Program at NASA Ames. The name of this organization was subsequently changed to the Numerical Aerospace Simulation Program, and more recently to the NASA Advanced Supercomputing Center, although the acronym remains 'NAS.' The developers of the original NPB suite were David H. Bailey, Eric Barszcz, John Barton, David Browning, Russell Carter, LeoDagum, Rod Fatoohi, Samuel Fineberg, Paul Frederickson, Thomas Lasinski, Rob Schreiber, Horst Simon, V. Venkatakrishnan and Sisira Weeratunga. The original NAS Parallel Benchmarks consisted of eight individual benchmark problems, each of which focused on some aspect of scientific computing. The principal focus was in computational aerophysics, although most of these benchmarks have much broader relevance, since in a much larger sense they are typical of many real-world scientific computing applications. The NPB suite grew out of the need for a more rational procedure to select new supercomputers for acquisition by NASA. The emergence of commercially available highly parallel computer systems in the late 1980s offered an attractive alternative to parallel vector supercomputers that had been the mainstay of high-end scientific computing. However, the introduction of highly parallel systems was accompanied by a regrettable level of hype, not only on the part of the commercial vendors but even, in some cases, by scientists using the systems. As a result, it was difficult to discern whether the new systems offered any fundamental performance advantage over vector supercomputers, and, if so, which of the parallel offerings would be most useful in real-world scientific computation. In part to draw attention to some of the performance reporting abuses prevalent at the time, the present author wrote a humorous essay 'Twelve Ways to Fool the Masses,' which described in a light-hearted way a number of the questionable ways in which both vendor marketing people and scientists were inflating and distorting their performance results. All of this underscored the need for an objective and scientifically defensible measure to compare performance on these systems.
Extremely large scale analysis is becoming increasingly important as supercomputers and their simulations move from petascale to exascale. The lack of dedicated hardware acceleration for rendering on today's supercomputing platforms motivates our detailed evaluation of the possibility of interactive rendering on the supercomputer. In order to facilitate our understanding of rendering on the supercomputing platform, we focus on scalability of rendering algorithms and architecture envisioned for exascale datasets. To understand tradeoffs for dealing with extremely large datasets, we compare three different rendering algorithms for large polygonal data: software based ray tracing, software based rasterization and hardware accelerated rasterization. We present a case study of strong and weak scaling of rendering extremely large data on both GPU and CPU based parallel supercomputers using Para View, a parallel visualization tool. Wc use three different data sets: two synthetic and one from a scientific application. At an extreme scale, algorithmic rendering choices make a difference and should be considered while approaching exascale computing, visualization, and analysis. We find software based ray-tracing offers a viable approach for scalable rendering of the projected future massive data sizes.
5G is more than an over-the-air radio technology upgrade. It is a strategy to extend Mobile Network Operator service offerings beyond traditional voice, instant messaging and Internet access. 5G Mobile Network Operators will offer new telecommunication services that include enhanced guarantees of confidentiality, integrity and availability. How could such services change the way Science collaborations connect scientists to supercomputers and other scientific facilities? Current scientific collaborations implicitly trust cloud service providers to securely store and process data. The perceived risks of outsourcing Science data security are counterbalanced by assurances that cloud providers operate at a scale that allows them to implement security measures impractical for Science collaborations (e.g. continuous system administrator behavioral monitoring and strict individual separation of duties). If that is true for a cloud service provider like Amazon Web Services (2018 revenue: $25.7 billion), could it also be true for Mobile Network Operators like Verizon Wireless (2018 revenue: $91.7 billion) or AT&T Mobility (2018 revenue: $71.3 billion)? DOE Leadership Class supercomputer facility users currently access them from the public Internet via Secure Shell. The sponsors and operators of the supercomputer facilities have determined that the public Internet path between the Scientist’s Device and the Login Node does not natively provide enough confidentiality or integrity to protect those communications. Therefore, the facilities achieve additional confidentiality and integrity by requiring Secure Shell encryption across those untrusted network paths. Using 5G Network Slice technology, a Mobile Network Operator may offer communication services between supercomputer users and facilities that natively provide confidentiality and integrity guarantees. Sponsors and operators of supercomputer facilities may determine that these guarantees provide enough confidentiality and integrity to protect those communications. If so, a 5G Network Slice could replace an SSH session running over the public Internet. Finally, this use case could be extended to other Office of Science user facility access requirements. Consider microscopy instruments at (e.g.) the Center for Nanoscale Materials or the Environmental Molecular Sciences Laboratory. The embedded systems controlling such instruments may not always support encrypted network access technologies like SSH. 5G Network Slices may offer an alternative to current VPN or SSH tunneling techniques, with additional benefits like guaranteed minimum bandwidth.
The Internet is a network of networks. It was originally funded by the Defense Advanced Research Projects Agency or DOD/DARPA and evolved in part from the connection of supercomputer sites across the United States. The National Science Foundation (NSF) made the most of their supercomputers by connecting the sites to each other. This made the supercomputers more efficient and now allows scientists, engineers and researchers to access the supercomputers from their own labs and offices. The high speed networks that connect the NSF supercomputers form the backbone of the Internet. The World Wide Web (WWW) is a menu system. It gathers Internet resources from all over the world into a series of screens that appear on your computer. The WWW is also a distributed. The distributed system stores data information on many computers (servers). These servers can go out and get data when you ask for it. Hypermedia is the base of the WWW. One can 'click' on a section and visit other hypermedia (pages). Our approach to demonstrating the importance of aviation research through the Internet began with learning how to put pages on the Internet (on-line) ourselves. We were assigned two aviation companies; Vision Micro Systems Inc. and Innovative Aerodynamic Technologies (IAT). We developed home pages for these SBIR companies. The equipment used to create the pages were the UNIX and Macintosh machines. HTML Supertext software was used to write the pages and the Sharp JX600S scanner to scan the images. As a result, with the use of the UNIX, Macintosh, Sun, PC, and AXIL machines, we were able to present our home pages to over 800,000 visitors.
An algorithm for decomposing large tally data in Monte Carlo particle transport simulations is developed, analyzed, and implemented in a continuous-energy Monte Carlo code, OpenMC. The algorithm is based on a non-overlapping decomposition of compute nodes into tracking processors and tally servers. The former are used to simulate the movement of particles through the domain while the latter continuously receive and update tally data. A performance model for this approach is developed, suggesting that, for a range of parameters relevant to LWR analysis, the tally server algorithm should perform with minimal overhead on contemporary supercomputers. An implementation of the algorithm in OpenMC is then tested on the Intrepid and Titan supercomputers, supporting the key predictions of the model over a wide range of parameters. We thus conclude that the tally server algorithm is a successful approach to circumventing classical on-node memory constraints en route to unprecedentedly detailed Monte Carlo reactor simulations.
Power consumption and associated carbon emissions are increasingly critical challenges for large-scale computing. Recent research proposes exploiting stranded power-uneconomic renewable power-for green supercomputing in a system called Zero-Carbon Cloud (ZCCloud). These efforts studied production supercomputing workloads on stranded-power based computing resources, demonstrating their achievable productivity. We explore economic viability of stranded-power based supercomputing, using three datacenter total-cost-of-ownership (TOO) models to study cost-effectiveness. These studies show that ZCCloud's approach can be cost-effective in the USA today, and is even more attractive in regions with higher power prices (e.g., Japan, Germany), achieving cost advantages as large as 50 percent. Environmental and power-grid benefits are a further advantage. We also explore the sensitivity of these results to changes in hardware TOO; cheaper hardware or longer lifetimes magnify the attractiveness of stranded-power based approaches, yielding advantages as large as 91 percent. These results are robust across different TCO models. Lastly, we study extreme-scale supercomputers ( >100 MW), finding stranded-power can increase peak capability per cost by as much as 80 percent.
In the race to PetaFLOP-speed supercomputing systems, the increase in computational capability has been accompanied by corresponding increases in CPU count, total RAM, and storage capacity. However, a proportional increase in storage bandwidth has lagged behind. In order to improve system reliability and to reduce maintenance effort for modern large-scale systems, system designers have opted to remove node-local storage from the compute nodes. Today's multi-TeraFLOP supercomputers are typically attached to parallel file systems that provide only tens of GBs/s of I/O bandwidth. As a result, such machines have access to much less than 1GB/s of I/O bandwidth per TeraFLOP of compute power, which is below the generally accepted limit required for a well-balanced system. In a many ways, the current I/O bottleneck limits the capabilities of modern supercomputers, specifically in terms of limiting their working sets and restricting fault tolerance techniques, which become critical on systems consisting of tens of thousands of components. This paper resolves the dilemma between high performance and high reliability by presenting an alternative system design which makes use of node-local storage to improve aggregate system I/O bandwidth. In this work, we focus on the checkpointing use-case and present an experimental evaluation of the Scalable Checkpoint/Restart (SCR) library, a new adaptive checkpointing library that uses node-local storage to significantly improve the checkpointing performance of large-scale supercomputers. Experiments show that SCR achieves unprecedented write speeds, reaching a measured 700GB/s of aggregate bandwidth on 8,752 processors and an estimated 1TB/s for a similarly structured machine of 12,500 processors. This corresponds to a speedup of over 70x compared to the bandwidth provided by the 10GB/s parallel file system the cluster uses. Further, SCR can adapt to an environment in which there is wide variation in performance or capacity among the individual node-local storage elements.
Summary Parallel I/O is a critical technique for moving data between compute and storage subsystems of supercomputers. With massive amounts of data produced or consumed by compute nodes, high‐performant parallel I/O is essential. I/O benchmarks play an important role in this process; however, there is a scarcity of I/O benchmarks representative of current workloads on HPC systems. Toward creating representative I/O kernels from real‐world applications, we have created h5bench , a set of I/O kernels that exercise hierarchical data format version 5 (HDF5) I/O on parallel file systems in numerous dimensions. Our focus on HDF5 is due to the parallel I/O library's heavy usage in various scientific applications running on supercomputing systems. The various tests benchmarked in the h5bench suite include I/O operations (read and write), data locality (arrays of basic data types and arrays of structures), array dimensionality (one‐dimensional arrays, two‐dimensional meshes, three‐dimensional cubes), I/O modes (synchronous and asynchronous). In this paper, we present the observed performance of h5bench executed along several of these dimensions on existing supercomputers (Cori and Summit) and pre‐exascale platforms (Perlmutter, Theta, and Polaris). h5bench measurements can be used to identify performance bottlenecks and their root causes and evaluate I/O optimizations. As the I/O patterns of h5bench are diverse and capture the I/O behaviors of various HPC applications, this study will be helpful to the broader supercomputing and I/O community.
Benchmarks for supercomputers are important tools, not only for evaluating and ranking modern supercomputers, but also for providing hints for future architecture design. As a new benchmark, HPGMG (high performance geometric multigrid) solves a linear equation set with a full geometric multi-grid algorithm. It involves computation on different scales, data movement with various volumes, global communication and neighbor communication with both large and small messages, etc., and is more correlated to real world applications than traditional benchmarks such as LINPACK. Therefore, it is desirable to examine how well HPGMG can perform on leadership supercomputers such as Sunway Taihulight. Sunway Taihulight, the No. 1 supercomputer in the Top 500 list from June 2016 to June 2018, which uses a specially designed many-core architecture SW26010, is of great interest to the community of high performance computing. With careful analysis and code design, we came up with an efficient implementation of HPGMG on SW26010 processors. We not only employed traditional optimization techniques such as 2.5D partitioning, double buffering, and collective data load, but also introduced a micro-benchmark to help with the choice of optimization direction and parameter tuning. Another contribution is that we proposed a new procedure for the major operations, by granulating and reordering the smooth function and the ghost exchange operation, leading to reduced memory copy and accelerated communication process. Our optimized implementation of HPGMG on Sunway TaihuLight achieved a ground-breaking performance of 1.036 × 10 12 Degrees of Freedom per second at the finest level, which is No. 1 on the HPGMG list of Nov 2017.