Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Message passing”

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.

At least 109 records · Page 6

Improving Multi-Million Virtual Rank MPI Execution in

(MUPI) is a parallel discrete event simulator designed for enabling software-based experimentation via simulated execution across a range of synthetic to unmodified parallel programs using the Message Passing Interface (MPI) with millions of tasks. Here, we report work in progress in improving the efficiency of . Among the issues uncovered are the scaling problems with implementing barriers and intertask message ordering. Preliminary performance shows the possibility of supporting hundreds of virtual MPI ranks per real processor core. Performance improvements of at least 2 are observed, and enable execution of benchmark MPI runs with over 16 million virtual ranks synchronized in a discrete event fashion on as few as 16,128 real cores of a Cray XT5.

99 GENERAL AND MISCELLANEOUS↗

Detection and Correction of Silent Data Corruption for Large-Scale High-Performance Computing

Faults have become the norm rather than the exception for high-end computing on clusters with 10s/100s of thousands of cores. Exacerbating this situation, some of these faults remain undetected, manifesting themselves as silent errors that corrupt memory while applications continue to operate and report incorrect results. This paper studies the potential for redundancy to both detect and correct soft errors in MPI message-passing applications. Our study investigates the challenges inherent to detecting soft errors within MPI application while providing transparent MPI redundancy. By assuming a model wherein corruption in application data manifests itself by producing differing MPI message data between replicas, we study the best suited protocols for detecting and correcting MPI data that is the result of corruption. To experimentally validate our proposed detection and correction protocols, we introduce RedMPI, an MPI library which resides in the MPI profiling layer. RedMPI is capable of both online detection and correction of soft errors that occur in MPI applications without requiring any modifications to the application source by utilizing either double or triple redundancy. Our results indicate that our most efficient consistency protocol can successfully protect applications experiencing even high rates of silent data corruption with runtime overheads between 0% and 30% as compared to unprotected applications without redundancy. Using our fault injector within RedMPI, we observe that even a single soft error can have profound effects on running applications, causing a cascading pattern of corruption in most cases causes that spreads to all other processes. RedMPI's protection has been shown to successfully mitigate the effects of soft errors while allowing applications to complete with correct results even in the face of errors.

Fiala, David J↗

Improving the Performance of the Extreme-scale Simulator

Investigating the performance of parallel applications at scale on future high-performance computing (HPC) architectures and the performance impact of different architecture choices is an important component of HPC hardware/software co-design. The Extreme-scale Simulator (xSim) is a simulation-based toolkit for investigating the performance of parallel applications at scale. xSim scales to millions of simulated Message Passing Interface (MPI) processes. The overhead introduced by a simulation tool is an important performance and productivity aspect. This paper documents two improvements to xSim: (1) a new deadlock resolution protocol to reduce the parallel discrete event simulation management overhead and (2) a new simulated MPI message matching algorithm to reduce the oversubscription management overhead. The results clearly show a significant performance improvement, such as by reducing the simulation overhead for running the NAS Parallel Benchmark suite inside the simulator from 1,020\% to 238% for the conjugate gradient (CG) benchmark and from 102% to 0% for the embarrassingly parallel (EP) and benchmark, as well as, from 37,511% to 13,808% for CG and from 3,332% to 204% for EP with accurate process failure simulation.

Engelmann, Christian↗

Evaluating MPI resource usage summary statistics

The Message Passing Interface (MPI) remains the dominant programming model for scientific applications running on today’s high-performance computing (HPC) systems. This dominance stems from MPI’s powerful semantics for inter-process communication that has enabled scientists to write applications for simulating important physical phenomena. MPI does not, however, specify how messages and synchronization should be carried out. Those details are typically dependent on low-level architecture details and the message characteristics of the application. Therefore, analyzing an application’s MPI resource usage is critical to tuning MPI’s performance on a particular platform. The result of this analysis is typically a discussion of the mean message sizes, queue search lengths and message arrival times for a workload or set of workloads. While a discussion of the arithmetic mean in MPI resource usage might be the most intuitive summary statistic, it is not always the most accurate in terms of representing the underlying data. In this paper, we analyze MPI resource usage for a number of key MPI workloads using an existing MPI trace collector and discrete-event simulator. Our analysis demonstrates that the average, while easy and efficient to calculate, is a useful metric for characterizing latency and bandwidth measurements, but may not be a good representation of application message sizes, match list search depths, or MPI inter-operation times. Additionally, we show that the median and mode are superior choices in many cases. We also observe that the arithmetic mean is not the best representation of central tendency for data that are drawn from distributions that are multi-modal or have heavy tails. Furthermore, the results and analysis of our work provide valuable guidance on how we, as a community, should discuss and analyze MPI resource usage data for scientific applications.

97 MATHEMATICS AND COMPUTING↗

A Fine-grained Asynchronous Bulk Synchronous parallelism model for PGAS applications

The Partitioned Global Address Space (PGAS) model is well suited for executing irregular applications on cluster-based systems, due to its efficient support for short, one-sided messages. Separately, the actor model has been gaining popularity as a productive asynchronous message-passing approach for distributed objects in enterprise and cloud computing platforms, typically implemented in languages such as Erlang, Scala or Rust. To the best of our knowledge, there has been no past work on using the actor model to deliver both productivity and scalability to irregular PGAS applications with large number of small messages. In this paper, we introduce a new programming system for PGAS applications, in which point-to-point remote operations can be expressed as fine-grained asynchronous actor messages. In our approach, the programmer does not need to worry about programming complexities related to message aggregation and termination detection. Our approach can be viewed as extending the classical Bulk Synchronous Parallelism model with fine-grained asynchronous communications within a phase or superstep. Here, we believe that our approach offers a desirable point in the productivity-performance space for PGAS applications, with more scalable performance and higher productivity relative to past approaches. Specifically, for seven irregular mini-applications from the Bale Kernels and three graph kernels executed using 2048 cores in the NERSC Cori system, our approach shows geometric mean performance improvements of ≥ 20X relative to standard PGAS versions (UPC and OpenSHMEM) while maintaining comparable productivity to those versions.

97 MATHEMATICS AND COMPUTING↗

RMA-MT Benchmarks

The RMA-MT benchmarks are a suite of performance tests for multi-threaded Message Passing Interface (MPI) Remote Memory Access (RMA) implementations. The suite includes tests for bandwidth and latency with any number of threads. This suite was originally developed and release by Sandia National Laboratory. LANL modifications to the suite include a complete re-write of much of the source code, bi-directional bandwidth, and message rate tests.

Hjelm, Nathan↗

UCX: An Open Source Framework for HPC Network APIs and Beyond, In: 2015 IEEE 23rd Annual Symposium on High-Performance Interconnects

This paper presents Unified Communication X (UCX), a set of network APIs and their implementations for high throughput computing. UCX comes from the combined effort of national laboratories, industry, and academia to design and implement a high-performing and highly-scalable network stack for next generation applications and systems. UCX design provides the ability to tailor its APIs and network functionality to suit a wide variety of application domains and hardware. We envision these APIs to satisfy the networking needs of many programming models such as Message Passing Interface (MPI), OpenSHMEM, Partitioned Global Address Space (PGAS) languages, task-based paradigms and I/O bound applications. To evaluate the design we implement the APIs and protocols, and measure the performance of overhead-critical network primitives fundamental for implementing many parallel programming models and system libraries. Our results show that the latency, bandwidth, and message rate achieved by the portable UCX prototype is very close to that of the underlying driver. With UCX, we achieved a message exchange latency of 0.89 us, a bandwidth of 6138.5 MB/s, and a message rate of 14 million messages per second. As far as we know, this is the highest bandwidth and message rate achieved by any network stack (publicly known) on this hardware.

Computer Science↗

High performance FPGA embedded system for machine learning based tracking and trigger in sPhenix and EIC

We present a comprehensive end-to-end pipeline to classify triggers versus background events in this paper. This pipeline makes online decisions to select signal data and enables the intelligent trigger system for efficient data collection in the Data Acquisition System (DAQ) of the upcoming sPHENIX and future EIC (Electron-Ion Collider) experiments. Starting from the coordinates of pixel hits that are lightened by passing particles in the detector, the pipeline applies three-stage of event processing (hits clustering, track reconstruction, and trigger detection) and labels all processed events with the binary tag of trigger versus background events. The pipeline consists of deterministic algorithms such as clustering pixels to reduce event size, tracking reconstruction to predict candidate edges, and advanced graph neural network-based models for recognizing the entire jet pattern. In particular, we apply the message-passing graph neural network to predict links between hits and reconstruct tracks and a hierarchical pooling algorithm (DiffPool) to make the graph-level trigger detection. We obtain an impressive performance (≥70% accuracy) for trigger detection with only 3200 neuron weights in the end-to-end pipeline. We deploy the end-to-end pipeline into a field-programmable gate array (FPGA) and accelerate the three stages with speedup factors of 1152, 280, and 21, respectively.

Instruments & Instrumentation↗

High-order finite-element seismic wave propagation modeling with MPI on a large GPU cluster

We implement a high-order finite-element application, which performs the numerical simulation of seismic wave propagation resulting for instance from earthquakes at the scale of a continent or from active seismic acquisition experiments in the oil industry, on a large cluster of NVIDIA Tesla graphics cards using the CUDA programming environment and non-blocking message passing based on MPI. Contrary to many finite-element implementations, ours is implemented successfully in single precision, maximizing the performance of current generation GPUs. We discuss the implementation and optimization of the code and compare it to an existing very optimized implementation in C language and MPI on a classical cluster of CPU nodes. We use mesh coloring to efficiently handle summation operations over degrees of freedom on an unstructured mesh, and non-blocking MPI messages in order to overlap the communications across the network and the data transfer to and from the device via PCIe with calculations on the GPU. We perform a number of numerical tests to validate the single-precision CUDA and MPI implementation and assess its accuracy. We then analyze performance measurements and depending on how the problem is mapped to the reference CPU cluster, we obtain a speedup of 20x or 12x.

97 MATHEMATICS AND COMPUTING↗

Redundant Execution of HPC Applications with MR-MPI

This paper presents a modular-redundant Message Passing Interface (MPI) solution, MR-MPI, for transparently executing high-performance computing (HPC) applications in a redundant fashion. The presented work addresses the deficiencies of recovery-oriented HPC, i.e., checkpoint/restart to/from a parallel file system, at extreme scale by adding the redundancy approach to the HPC resilience portfolio. It utilizes the MPI performance tool interface, PMPI, to transparently intercept MPI calls from an application and to hide all redundancy-related mechanisms. A redundantly executed application runs with r*m native MPI processes, where r is the number of MPI ranks visible to the application and m is the replication degree. Messages between redundant nodes are replicated. Partial replication for tunable resilience is supported. The performance results clearly show the negative impact of the O(m^m) messages between replicas. For low-level, point-to-point benchmarks, the impact can be as high as the replication degree. For applications, performance highly depends on the actual communication types and counts. On single-core systems, the overhead can be 0% for embarrassingly parallel applications independent of the employed redundancy configuration or up to 70-90% for communication-intensive applications in a dual-redundant configuration. On multi-core systems, the overhead can be significantly higher due to the additional communication contention.

Engelmann, Christian↗

Building a Fault Tolerant MPI Application: A Ring Communication Example

Process failure is projected to become a normal event for many long running and scalable High Performance Computing (HPC) applications. As such many application developers are investigating Algorithm Based Fault Tolerance (ABFT) techniques to improve the efficiency of application recovery beyond what existing checkpoint/restart techniques alone can provide. Unfortunately for these application developers the libraries that their applications depend upon, like Message Passing Interface (MPI), do not have standardized fault tolerance semantics. This paper introduces the reader to a set of run-through stabilization semantics being developed by the MPI Forum's Fault Tolerance Working Group to support ABFT. Using a well-known ring communication program as the running example, this paper illustrates to application developers new to ABFT some of the issues that arise when designing a fault tolerant application. The ring program allows the paper to focus on the communication-level issues rather than the data preservation mechanisms covered by existing literature. This paper highlights a common set of issues that application developers must address in their design including program control management, duplicate message detection, termination detection, and testing. The discussion provides application developers new to ABFT with an introduction to both new interfaces becoming available, and a range of design issues that they will likely need to address regardless of their research domain.

99 GENERAL AND MISCELLANEOUS↗

Optimizing blocking and nonblocking reduction operations for multicore systems: Hierarchical design and implementation

Many scientific simulations, using the Message Passing Interface (MPI) programming model, are sensitive to the performance and scalability of reduction collective operations such as MPI_Allreduce and MPI_Reduce. These operations are the most widely used abstractions to perform mathematical operations over all processes that are part of the simulation. In this work, we propose a hierarchical design to implement the reduction operations on multicore systems. This design aims to improve the efficiency of reductions by 1) tailoring the algorithms and customizing the implementations for various communication mechanisms in the system 2) providing the ability to configure the depth of hierarchy to match the system architecture, and 3) providing the ability to independently progress each of this hierarchy. Using this design, we implement MPI_Allreduce and MPI_Reduce operations (and its nonblocking variants MPI_Iallreduce and MPI_Ireduce) for all message sizes, and evaluate on multiple architectures including InfiniBand and Cray XT5. We leverage and enhance our existing infrastructure, Cheetah, which is a framework for implementing hierarchical collective operations to implement these reductions. The experimental results show that the Cheetah reduction operations outperform the production-grade MPI implementations such as Open MPI default, Cray MPI, and MVAPICH2, demonstrating its efficiency, flexibility and portability. On Infini-Band systems, with a microbenchmark, a 512-process Cheetah nonblocking Allreduce and Reduce achieves a speedup of 23x and 10x, respectively, compared to the default Open MPI reductions. The blocking variants of the reduction operations also show similar performance benefits. A 512-process nonblocking Cheetah Allreduce achieves a speedup of 3x, compared to the default MVAPICH2 Allreduce implementation. On a Cray XT5 system, a 6144-process Cheetah Allreduce outperforms the Cray MPI by 145%. The evaluation with an application kernel, Conjugate Gradient solver, shows that the Cheetah reductions speeds up total time to solution by 195%, demonstrating the potential benefits for scientific simulations.

Venkata, Manjunath Gorentla↗

Optimizing Blocking and Nonblocking Reduction Operations for Multicore Systems: Hierarchical Design and Implementation

Many scientific simulations, using the Message Passing Interface (MPI) programming model, are sensitive to the performance and scalability of reduction collective operations such as MPI Allreduce and MPI Reduce. These operations are the most widely used abstractions to perform mathematical operations over all processes that are part of the simulation. In this work, we propose a hierarchical design to implement the reduction operations on multicore systems. This design aims to improve the efficiency of reductions by 1) tailoring the algorithms and customizing the implementations for various communication mechanisms in the system 2) providing the ability to configure the depth of hierarchy to match the system architecture, and 3) providing the ability to independently progress each of this hierarchy. Using this design, we implement MPI Allreduce and MPI Reduce operations (and its nonblocking variants MPI Iallreduce and MPI Ireduce) for all message sizes, and evaluate on multiple architectures including InfiniBand and Cray XT5. We leverage and enhance our existing infrastructure, Cheetah, which is a framework for implementing hierarchical collective operations to implement these reductions. The experimental results show that the Cheetah reduction operations outperform the production-grade MPI implementations such as Open MPI default, Cray MPI, and MVAPICH2, demonstrating its efficiency, flexibility and portability. On Infini- Band systems, with a microbenchmark, a 512-process Cheetah nonblocking Allreduce and Reduce achieves a speedup of 23x and 10x, respectively, compared to the default Open MPI reductions. The blocking variants of the reduction operations also show similar performance benefits. A 512-process nonblocking Cheetah Allreduce achieves a speedup of 3x, compared to the default MVAPICH2 Allreduce implementation. On a Cray XT5 system, a 6144-process Cheetah Allreduce outperforms the Cray MPI by 145%. The evaluation with an application kernel, Conjugate Gradient solver, shows that the Cheetah reductions speeds up total time to solution by 195%, demonstrating the potential benefits for scientific simulations.

Gorentla Venkata, Manjunath↗

Designing Energy Efficient Communication Runtime Systems for Data Centric Programming Models

The insatiable demand of high performance computing is being driven by the most computationally intensive applications such as computational chemistry, climate modeling, nuclear physics, etc. The last couple of decades have observed a tremendous rise in supercomputers with architectures ranging from traditional clusters to system-on-a-chip in order to achieve the petaflop computing barrier. However, with advent of petaflop-plus computing, we have ushered in an era where power efficient system software stack is imperative for execution on exascale systems and beyond. At the same time, computationally intensive applications are exploring programming models beyond traditional message passing, as a combination of Partitioned Global Address Space (PGAS) languages and libraries, providing one-sided communication paradigm with put, get and accumulate primitives. To support the PGAS models, it is critical to design power efficient and high performance one-sided communication runtime system. In this paper, we design and implement PASCoL, a high performance power aware one-sided communication library using Aggregate Remote Memory Copy Interface (ARMCI), the communication runtime system of Global Arrays. For various communication primitives provided by ARMCI, we study the impact of Dynamic Voltage/Frequency Scaling (DVFS) and a combination of interrupt (blocking)/polling based mechanisms provided by most modern interconnects. We implement our design and evaluate it with synthetic benchmarks using an InfiniBand cluster. Our results indicate that PASCoL can achieve significant reduction in energy consumed per byte transfer without additional penalty for various one-sided communication primitives and various message sizes and data transfer patterns.

99 GENERAL AND MISCELLANEOUS↗

Portable and Transparent Message Compression in MPI Libraries to Improve the Performance and Scalability of Parallel Applications

The goal of this project has been to develop a lossless compression algorithm for message-passing libraries that can accelerate HPC systems by reducing the communication time. Because both compression and decompression have to be performed in software in real time, the algorithm has to be extremely fast while still delivering a good compression ratio. During the first half of this project, they designed a new compression algorithm called FPC for scientific double-precision data, made the source code available on the web, and published two papers describing its operation, the first in the proceedings of the Data Compression Conference and the second in the IEEE Transactions on Computers. At comparable average compression ratios, this algorithm compresses and decompresses 10 to 100 times faster than BZIP2, DFCM, FSD, GZIP, and PLMI on the three architectures tested. With prediction tables that fit into the CPU's L1 data acache, FPC delivers a guaranteed throughput of six gigabits per second on a 1.6 GHz Itanium 2 system. The C source code and documentation of FPC are posted on-line and have already been downloaded hundreds of times. To evaluate FPC, they gathered 13 real-world scientific datasets from around the globe, including satellite data, crash-simulation data, and messages from HPC systems. Based on the large number of requests they received, they also made these datasets available to the community (with permission of the original sources). While FPC represents a great step forward, it soon became clear that its throughput was too slow for the emerging 10 gigabits per second networks. Hence, no speedup can be gained by including this algorithm in an MPI library. They therefore changed the aim of the second half of the project. Instead of implementing FPC in an MPI library, they refocused their efforts to develop a parallel compression algorithm to further boost the throughput. After all, all modern high-end microprocessors contain multiple CPUs on a single die and will undoubtedly contain even larger numbers of CPUs in the future. Hence, a compute node in a parallel system is likely to have multiple processors per interconnection link. Thus, they investigated how to best exploit multiple CPUs to compress and decompress data for a single link, i.e., they developed a parallel version of the FPC algorithm called pFPC. On a 4-core 3 GHz Xeon system, pFPC exceeds 10 gigabits per second both in compression and decompression throughput while still reaching a compression ratio of 18%, which is reasonable for their double-precision floating-point datasets. At half this throughput, it achieves a 41% compression ratio. Moreover, the algorithm scales well to four cores and promises even higher throughputs on larger machines. Additional results and a description of the pFPC algorithm have recently been published in the proceedings of the Data Compression Conference. The C source code of pFPC is also publicly available.

97 MATHEMATICS AND COMPUTING↗

Comparison of DeePMD, MTP, GAP, ACE and MACE Machine‐Learned Potentials for Radiation‐Damage Simulations: A User Perspective

Accurate and efficient interatomic potentials are essential for molecular dynamics (MD) simulations of radiation damage, gas diffusion, and phase stability in complex ceramics such as LiAlO 2 , especially under extreme conditions relevant to tritium production. Here, we evaluate the performance of six machine-learned interatomic potentials (MLIPs), moment tensor potential (MTP), Gaussian approximation potential, deep potential (DeePMD), atomic cluster expansion (ACE), message-passing ACE (multilayer atomic cluster expansion (MACE) pretrained) and MACE (trained from-scratch), all trained on the same density functional theory dataset with inclusion of tritium. The MLIPs are benchmarked against traditional Buckingham and ReaxFF potentials in terms of energy accuracy, density predictions, thermal equilibration behavior, threshold displacement energy (E d ), tritium diffusivity, and computational cost. Among the models, MTP shows the best overall balance between efficiency and accuracy, with low force and energy errors and realistic E d values for Li and Al. The ACE and MACE (pretrained and trained from scratch) models exhibit high E d (>200 eV) and unphysical pair interactions. DeePMD underestimates Ed due to overly repulsive behavior even at equilibrium distances. All models over-estimate tritium diffusion but the pretrained MACE model behaves well during tritium-diffusion simulations up to 500 K, maintaining diffusivities in the physically consistent 10 −11 m 2 /s range. Finally, we quantify the computational cost of each potential in large-scale atomic/molecular massively parallel simulator, finding that only MTP is more efficient than traditional empirical potentials, while others are significantly more expensive. These findings explain the trade-offs between accuracy and computational cost in MLIP development and provide essential guidance for use in high-throughput radiation damage and gas diffusion simulations in nuclear ceramics.

74 ATOMIC AND MOLECULAR PHYSICS↗

Clock Synchronization in High-end Computing Environments: A Strategy for Minimizing Clock Variance at Runtime

We present a new software-based clock synchronization scheme that provides high precision time agreement among distributed memory nodes. The technique is designed to minimize variance from a reference chimer during runtime and with minimal time-request latency. Our scheme permits initial unbounded variations in time and corrects both slow and fast chimers (clock skew). An implementation developed within the context of the MPI message passing interface is described, and time coordination measurements are presented. Among our results, the mean time variance for a set of nodes improved from 20.0 milliseconds under standard Network Time Protocol (NTP) down to 2.29 secs under our scheme.

clock synchronization↗

Performance analysis of distributed symmetric sparse matrix vector multiplication algorithm for multi-core architectures

In this article, sparse matrix vector multiply (SpMVM) is an important kernel that frequently arises in high performance computing applications. Due to its low arithmetic intensity, several approaches have been proposed in literature to improve its scalability and efficiency in large scale computations. In this paper, our target systems are high end multi-core architectures and we use messaging passing interface + open multiprocessing hybrid programming model for parallelism. We analyze the performance of recently proposed implementation of the distributed symmetric SpMVM, originally developed for large sparse symmetric matrices arising in ab initio nuclear structure calculations. We also study important features of this implementation and compare with previously reported implementations that do not exploit underlying symmetry. Our SpMVM implementations leverage the hybrid paradigm to efficiently overlap expensive communications with computations. Our main comparison criterion is the "CPU core hours" metric, which is the main measure of resource usage on supercomputers. We analyze the effects of topology-aware mapping heuristic using simplified network load model. Furthermore, we have tested the different SpMVM implementations on two large clusters with 3D Torus and Dragonfly topology. Our results show that the distributed SpMVM implementation that exploits matrix symmetry and hides communication yields the best value for the "CPU core hours" metric and significantly reduces data movement overheads.

97 MATHEMATICS AND COMPUTING↗