Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Bandwidth”

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 955 records · Page 53

Lossy checkpoint compression in full waveform inversion: a case study with ZFPv0.5.5 and the overthrust model

This paper proposes a new method that combines checkpointing methods with error-controlled lossy compression for large-scale high-performance full-waveform inversion (FWI), an inverse problem commonly used in geophysical exploration. This combination can significantly reduce data movement, allowing a reduction in run time as well as peak memory. In the exascale computing era, frequent data transfer (e.g., memory bandwidth, PCIe bandwidth for GPUs, or network) is the performance bottleneck rather than the peak FLOPS of the processing unit. Like many other adjoint-based optimization problems, FWI is costly in terms of the number of floating-point operations, large memory footprint during backpropagation, and data transfer overheads. Past work for adjoint methods has developed checkpointing methods that reduce the peak memory requirements during backpropagation at the cost of additional floating-point computations. Combining this traditional checkpointing with error-controlled lossy compression, we explore the three-way tradeoff between memory, precision, and time to solution. We investigate how approximation errors introduced by lossy compression of the forward solution impact the objective function gradient and final inverted solution. Empirical results from these numerical experiments indicate that high lossy-compression rates (compression factors ranging up to 100) have a relatively minor impact on convergence rates and the quality of the final solution.

58 GEOSCIENCES↗

Refactoring the elastic–viscous–plastic solver from the sea ice model CICE v6.5.1 for improved performance

This study focuses on the performance of the elastic–viscous–plastic (EVP) dynamical solver within the sea ice model, CICE v6.5.1. The study has been conducted in two steps. First, the standard EVP solver was extracted from CICE for experiments with refactored versions, which are used for performance testing. Second, one refactored version was integrated and tested in the full CICE model to demonstrate that the new algorithms do not significantly impact the physical results. The study reveals two dominant bottlenecks, namely (1) the number of Message Parsing Interface (MPI) and Open Multi-Processing (OpenMP) synchronization points required for halo exchanges during each time step combined with the irregular domain of active sea ice points and (2) the lack of single-instruction, multiple-data (SIMD) code generation. The standard EVP solver has been refactored based on two generic patterns. The first pattern exposes how general finite differences on masked multi-dimensional arrays can be expressed in order to produce significantly better code generation by changing the memory access pattern from random access to direct access. The second pattern takes an alternative approach to handle static grid properties. The measured single-core performance improvement is more than a factor of 5 compared to the standard implementation. The refactored implementation of strong scales on the Intel® Xeon® Scalable Processors series node until the available bandwidth of the node is used. For the Intel® Xeon® CPU Max series, there is sufficient bandwidth to allow the strong scaling to continue for all the cores on the node, resulting in a single-node improvement factor of 35 over the standard implementation. This study also demonstrates improved performance on GPU processors.

58 GEOSCIENCES↗

System Modeling of kJ-class Petawatt Lasers at LLNL

Advanced Radiographic Capability (ARC) project at the National Ignition Facility (NIF) is designed to produce energetic, ultrafast x-rays in the range of 70-100 keV for backlighting NIF targets. The chirped pulse amplification (CPA) laser system will deliver kilo-Joule pulses at an adjustable pulse duration from 1 ps to 50 ps. System complexity requires sophisticated simulation and modeling tools for design, performance prediction, and comprehension of experimental results. We provide a brief overview of ARC, present our main modeling tools, and describe important performance predictions. The laser system (Fig. 1) consists of an all-fiber front end, including chirped-fiber Bragg grating (CFBG) stretchers. The beam after the final fiber amplifier is split into two apertures and spatially shaped. The split beam first seeds a regenerative amplifier and is then amplified in a multi-pass Nd:glass amplifier. Next, the preamplified chirped pulse is split in time into four identical replicas and injected into one NIF Quad. At the output of the NIF beamline, each of the eight amplified pulses is compressed in an individual, folded, four-grating compressor. Compressor grating pairs have slightly different groove densities to enable compact folding geometry and eliminate adjacent beam cross-talk. Pulse duration is adjustable with a small, rack-mounted compressor in the front-end. We use non-sequential ray-tracing software, FRED for design and layout of the optical system. Currently, our FRED model includes all of the optical components from the output of the fiber front end to the target center (Fig. 2). CAD designed opto-mechanical components are imported into our FRED model to provide a complete system description. In addition to incoherent ray tracing and scattering analysis, FRED uses Gaussian beam decomposition to model coherent beam propagation. Neglecting nonlinear effects, we can obtain a nearly complete frequency domain description of the ARC beam at different stages in the system. We employ 3D Fourier based propagation codes: MIRO, Virtual Beamline (VBL), and PROP for time-domain pulse analysis. These codes simulate nonlinear effects, calculate near and far field beam profiles, and account for amplifier gain. Verification of correct system set-up is a major difficulty to using these codes. VBL and PROP predictions have been extensively benchmarked to NIF experiments, and the verified descriptions of specific NIF beamlines are used for ARC. MIRO has the added capability of treating bandwidth specific effects of CPA. A sample MIRO model of the NIF beamline is shown in Fig. 3. MIRO models are benchmarked to VBL and PROP in the narrow bandwidth mode. Developing a variety of simulation tools allows us to cross-check predictions of different models and gain confidence in their fidelity. Preliminary experiments, currently in progress, are allowing us to validate and refine our models, and help guide future experimental campaigns.

42 ENGINEERING↗

Theoretical analysis of acceleration measurements in a model of an operating wind turbine.

Wind loading from turbulence and gusts can cause damage in horizontal axis wind turbines. These unsteady loads and the resulting damage initiation and propagation are difficult to predict. Unsteady loads enter at the rotor and are transmitted to the drivetrain. The current generation of wind turbine has drivetrain-mounted vibration and bearing temperature sensors, a nacelle-mounted inertial measurement unit, and a nacelle-mounted anemometer and wind vane. Some advanced wind turbines are also equipped with strain measurements at the root of the rotor. This paper analyzes additional measurements in a rotor blade to investigate the complexity of these unsteady loads. By identifying the spatial distribution, amplitude, and frequency bandwidth of these loads, design improvements could be facilitated to reduce uncertainties in reliability predictions. In addition, dynamic load estimates could be used in the future to control high-bandwidth aerodynamic actuators distributed along the rotor blade to reduce the saturation of slower pitch actuators currently used for wind turbine blades. Local acceleration measurements are made along a rotor blade to infer operational rotor states including deflection and dynamic modal contributions. Previous work has demonstrated that acceleration measurements can be experimentally acquired on an operating wind turbine. Simulations on simplified rotor blades have also been used to demonstrate that mean blade loading can be estimated based on deflection estimates. To successfully apply accelerometers in wind turbine applications for load identification, the spectral and spatial characteristics of each excitation source must be understood so that the total acceleration measurement can be decomposed into contributions from each source. To demonstrate the decomposition of acceleration measurements in conjunction with load estimation methods, a flexible body model has been created with MSC.ADAMS{copyright} The benefit of using a simulation model as opposed to a physical experiment to examine the merits of acceleration-based load identification methods is that models of the structural dynamics and aerodynamics enable one to compare estimates of the deflection and loading with actual values. Realistic wind conditions are applied to the wind turbine and used to estimate the operational displacement and acceleration of the rotor. The per-revolution harmonics dominate the displacement and acceleration response. Turbulent wind produces broadband excitation that includes both the harmonics and modal vibrations, such as the tower modes. Power Spectral Density estimates of the acceleration along the span of the rotor blades indicate that the edge modes may be coupled to the second harmonic.

17 WIND ENERGY↗

Calibration of the modulation transfer function of surface profilometers with binary pseudo-random test standards: expanding the application range

A modulation transfer function (MTF) calibration method based on binary pseudo-random (BPR) gratings and arrays [Proc. SPIE 7077-7 (2007), Opt. Eng. 47, 073602 (2008)] has been proven to be an effective MTF calibration method for a number of interferometric microscopes and a scatterometer [Nucl. Instr. and Meth. A616, 172 (2010)]. Here we report on a further expansion of the application range of the method. We describe the MTF calibration of a 6 inch phase shifting Fizeau interferometer. Beyond providing a direct measurement of the interferometer's MTF, tests with a BPR array surface have revealed an asymmetry in the instrument's data processing algorithm that fundamentally limits its bandwidth. Moreover, the tests have illustrated the effects of the instrument's detrending and filtering procedures on power spectral density measurements. The details of the development of a BPR test sample suitable for calibration of scanning and transmission electron microscopes are also presented. Such a test sample is realized as a multilayer structure with the layer thicknesses of two materials corresponding to BPR sequence. The investigations confirm the universal character of the method that makes it applicable to a large variety of metrology instrumentation with spatial wavelength bandwidths from a few nanometers to hundreds of millimeters.

42 ENGINEERING↗

Diagnosing ICF gamma-ray physics

Gamma rays produced in an ICF environment open up a host of physics opportunities we are just beginning to explore. A branch of the DT fusion reaction, with a branching ratio on the order of 2e-5 {gamma}/n, produces 16.7 MeV {gamma}-rays. These {gamma}-rays provide a direct measure of fusion reaction rate (unlike x-rays) without being compromised by Doppler spreading (unlike neutrons). Reaction-rate history measurements, such as nuclear bang time and burn width, are fundamental quantities that will be used to optimize ignition on the National Ignition Facility (NIF). Gas Cherenkov Detectors (GCD) that convert fusion {gamma}-rays to UV/visible Cherenkov photons for collection by fast optical recording systems established their usefulness in illuminating ICF physics in several experimental campaigns at OMEGA. Demonstrated absolute timing calibrations allow bang time measurements with accuracy better than 30 ps. System impulse response better than 95 ps fwhm have been made possible by the combination of low temporal dispersion GCDs, ultra-fast microchannel-plate photomultiplier tubes (PMT), and high-bandwidth Mach Zehnder fiber optic data links and digitizers, resulting in burn width measurement accuracy better than 10ps. Inherent variable energy-thresholding capability allows use of GCDs as {gamma}-ray spectrometers to explore other interesting nuclear processes. Recent measurements of the 4.44 MeV {sup 12}C(n,n{prime}) {gamma}-rays produced as 14.1 MeV DT fusion neutrons pass through plastic capsules is paving the way for a new CH ablator areal density measurement. Insertion of various neutron target materials near target chamber center (TCC) producing secondary, neutron-induced {gamma}y-rays are being used to study other nuclear interactions and as in-situ sources to calibrate detector response and DT branching ratio. NIF Gamma Reaction History (GRH) diagnostics, based on the GCD concept, are now being developed based on optimization of sensitivity, bandwidth, dynamic range, cost, and NIF-specific logistics, requirements and extreme radiation environment. Implementation will occur in two phases: (1) four PMT-based channels mounted to the outside of the target chamber at {approx}6m from TCC (GRH-6m) for the 3e13-3e16 DT neutron yield range expected during the early ignition-tuning campaigns; and (2) several channels located just inside the target bay shield wall at 15 m from TCC (GRH-15m) with optical paths leading through the cement shield wall to well-shielded streak cameras and PMTs for the 1e16-1e20 yield range expected during the DT ignition campaign. Multiple channels at each phase will allow for increased redundancy, reliability, accuracy and flexibility. This suite of diagnostics will make possible exploration of interesting {gamma}-ray physics well beyond the ignition campaign.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

End-to-End Network QoS via Scheduling of Flexible Resource Reservation Requests

Modern data-intensive applications move vast amounts of data between multiple locations around the world. To enable predictable and reliable data transfer, next generation networks allow such applications to reserve network resources for exclusive use. In this paper, we solve an important problem (called SMR3) to accommodate multiple and concurrent network reservation requests between a pair of end-sites. Given the varying availability of bandwidth within the network, our goal is to accommodate as many reservation requests as possible while minimizing the total time needed to complete the data transfers. We first prove that SMR3 is an NP-hard problem. Then we solve it by developing a polynomial-time heuristic, called RRA. The RRA algorithm hinges on an efficient mechanism to accommodate large number of requests by minimizing the bandwidth wastage. Finally, via numerical results, we show that RRA constructs schedules that accommodate significantly larger number of requests compared to other, seemingly efficient, heuristics.

97 MATHEMATICS AND COMPUTING↗

Integration experiments and performance studies of a COTS parallel archive system

Current and future Archive Storage Systems have been asked to (a) scale to very high bandwidths, (b) scale in metadata performance, (c) support policy-based hierarchical storage management capability, (d) scale in supporting changing needs of very large data sets, (e) support standard interface, and (f) utilize commercial-off-the-shelf (COTS) hardware. Parallel file systems have been asked to do the same thing but at one or more orders of magnitude faster in performance. Archive systems continue to move closer to file systems in their design due to the need for speed and bandwidth, especially metadata searching speeds such as more caching and less robust semantics. Currently the number of extreme highly scalable parallel archive solutions is very small especially those that will move a single large striped parallel disk file onto many tapes in parallel. We believe that a hybrid storage approach of using COTS components and innovative software technology can bring new capabilities into a production environment for the HPC community much faster than the approach of creating and maintaining a complete end-to-end unique parallel archive software solution. In this paper, we relay our experience of integrating a global parallel file system and a standard backup/archive product with a very small amount of additional code to provide a scalable, parallel archive. Our solution has a high degree of overlap with current parallel archive products including (a) doing parallel movement to/from tape for a single large parallel file, (b) hierarchical storage management, (c) ILM features, (d) high volume (non-single parallel file) archives for backup/archive/content management, and (e) leveraging all free file movement tools in Linux such as copy, move, Is, tar, etc. We have successfully applied our working COTS Parallel Archive System to the current world's first petafiop/s computing system, LANL's Roadrunner machine, and demonstrated its capability to address requirements of future archival storage systems.

99 GENERAL AND MISCELLANEOUS↗

Laser System for Livermore's Mono Energetic Gamma-Ray Source

A Mono-energetic Gamma-ray (MEGa-ray) source, based on Compton scattering of a high-intensity laser beam off a highly relativistic electron beam, requires highly specialized laser systems. To minimize the bandwidth of the {gamma}-ray beam, the scattering laser must have minimal bandwidth, but also match the electron beam depth of focus in length. This requires a {approx}1 J, 10 ps, fourier-transform-limited laser system. Also required is a high-brightness electron beam, best provided by a photoinjector. This electron source requires a second laser system with stringent requirements on the beam including flat transverse and longitudinal profiles and fast rise times. Furthermore, these systems must be synchronized to each other with ps-scale accuracy. Using a novel hyper-dispersion compressor configuration and advanced fiber amplifiers and diode-pumped Nd:YAG amplifiers, we have designed laser systems that meet these challenges for the X-band photoinjector and Compton-scattering source being built at Lawrence Livermore National Laboratory.

42 ENGINEERING↗

Beam Position Monitor System for PEP II

We describe the beam position monitor system built for PEP-II, the B-factory at SLAC. The system reports beam position for bunches of between 5 x 10{sup 8} and 8 x 10{sup 10} electron charges, either singly or as continuous streams of bunches every 4.2 ns. Resolution at full charge is to be better than 10 microns in a single turn. Higher resolution is available via on-board multi-turn averaging. The position signal is processed in a 20 MHz bandwidth around 952 MHz. This bandwidth, rather broader than that typical of RF position monitors, allows good resolution for low charge single bunches. Additional novel features include stringent control of return losses in order to minimize cross-talk between nearby bunches which may contain very different charges. The digitizing electronics is multiplexed between the two PEP-II storage rings. Design, construction, and installation experience, as well as first results with beam are presented.

43 PARTICLE ACCELERATORS↗

Enhancing I/O Throughput via Efficient Routing and Placement for Large-scale Parallel File Systems

As storage systems get larger to meet the demands of petascale systems, careful planning must be applied to avoid congestion points and extract the maximum performance. In addition, the large data sets generated by such systems makes it desirable for all compute resources to have common access to this data without needing to copy it to each machine. This paper describes a method of placing I/O close to the storage nodes to minimize contention on Cray s SeaStar2+ network, and extends it to a routed Lustre configuration to gain the same benefits when running against a center-wide file system. Our experiments using half of the resources of Spider the center-wide file system at the Oak Ridge Leadership Computing Facility show that I/O write bandwidth can be improved by up to 45% (from 71.9 to 104 GB/s) for a direct-attached configuration and by 137% (47.6 GB/s to 115 GB/s) for a routed configuration. We demonstrated up to 20.7% reduction in run-time for production scientific applications. With the full Spider system, we demonstrated over 240 GB/s of aggregate bandwidth using our techniques.

99 GENERAL AND MISCELLANEOUS↗

Active Flash: Performance-Energy Tradeoffs for Out-of-Core Processing on Non-Volatile Memory Devices

In this abstract, we study the performance and energy tradeoffs involved in migrating data analysis into the flash device, a process we refer to as Active Flash. The Active Flash paradigm is similar to 'active disks', which has received considerable attention. Active Flash allows us to move processing closer to data, thereby minimizing data movement costs and reducing power consumption. It enables true out-of-core computation. The conventional definition of out-of-core solvers refers to an approach to process data that is too large to fit in the main memory and, consequently, requires access to disk. However, in Active Flash, processing outside the host CPU literally frees the core and achieves real 'out-of-core' analysis. Moving analysis to data has long been desirable, not just at this level, but at all levels of the system hierarchy. However, this requires a detailed study on the tradeoffs involved in achieving analysis turnaround under an acceptable energy envelope. To this end, we first need to evaluate if there is enough computing power on the flash device to warrant such an exploration. Flash processors require decent computing power to run the internal logic pertaining to the Flash Translation Layer (FTL), which is responsible for operations such as address translation, garbage collection (GC) and wear-leveling. Modern SSDs are composed of multiple packages and several flash chips within a package. The packages are connected using multiple I/O channels to offer high I/O bandwidth. SSD computing power is also expected to be high enough to exploit such inherent internal parallelism within the drive to increase the bandwidth and to handle fast I/O requests. More recently, SSD devices are being equipped with powerful processing units and are even embedded with multicore CPUs (e.g. ARM Cortex-A9 embedded processor is advertised to reach 2GHz frequency and deliver 5000 DMIPS; OCZ RevoDrive X2 SSD has 4 SandForce controllers, each with 780MHz max frequency Tensilica core). Efforts that take advantage of the available computing cycles on the processors on SSDs to run auxiliary tasks other than actual I/O requests are beginning to emerge. Kim et al. investigate database scan operations in the context of processing on the SSDs, and propose dedicated hardware logic to speed up scans. Also, cluster architectures have been explored, which consist of low-power embedded CPUs coupled with small local flash to achieve fast, parallel access to data. Processor utilization on SSD is highly dependent on workloads and, therefore, they can be idle during periods with no I/O accesses. We propose to use the available processing capability on the SSD to run tasks that can be offloaded from the host. This paper makes the following contributions: (1) We have investigated Active Flash and its potential to optimize the total energy cost, including power consumption on the host and the flash device; (2) We have developed analytical models to analyze the performance-energy tradeoffs for Active Flash, by treating the SSD as a blackbox, this is particularly valuable due to the proprietary nature of the SSD internal hardware; and (3) We have enhanced a well-known SSD simulator (from MSR) to implement 'on-the-fly' data compression using Active Flash. Our results provide a window into striking a balance between energy consumption and application performance.

42 ENGINEERING↗

Performance study of GPUs in real-time trigger applications for HEP experiments

Graphical Processing Units (GPUs) have evolved into highly parallel, multi-threaded, multicore powerful processors with high memory bandwidth. GPUs are used in a variety of intensive computing applications. The combination of highly parallel architecture and high memory bandwidth makes GPUs a potentially promising technology for effective real-time processing for High Energy Physics (HEP) experiments. However, not much is known of their performance in real-time applications that require low latency, such as the trigger for HEP experiments. We describe an R and D project with the goal to study the performance of GPU technology for possible low latency applications, performing basic operations as well as some more advanced HEP lower-level trigger algorithms (such as fast tracking or jet finding). We present some preliminary results on timing measurements, comparing the performance of a CPU versus a GPU with NVIDIA's CUDA general-purpose parallel computing architecture, carried out at CDF's Level-2 trigger test stand. These studies will provide performance benchmarks for future studies to investigate the potential and limitations of GPUs for real-time applications in HEP experiments.

43 PARTICLE ACCELERATORS↗

LIDAR Wind Speed Measurements of Evolving Wind Fields

Light Detection and Ranging (LIDAR) systems are able to measure the speed of incoming wind before it interacts with a wind turbine rotor. These preview wind measurements can be used in feedforward control systems that are designed to reduce turbine loads. However, the degree to which such preview-based control techniques can reduce loads by reacting to turbulence depends on how accurately the incoming wind field can be measured. Past studies have assumed the validity of physicist G.I. Taylor's 1938 frozen turbulence hypothesis, which implies that turbulence remains unchanged as it advects downwind at the mean wind speed. With Taylor's hypothesis applied, the only source of wind speed measurement error is distortion caused by the LIDAR. This study introduces wind evolution, characterized by the longitudinal coherence of the wind, to LIDAR measurement simulations using the National Renewable Energy Laboratory's (NREL's) 5-megawatt turbine model to create a more realistic measurement model. A simple model of wind evolution was applied to a frozen wind field that was used in previous studies to investigate the effects of varying the intensity of wind evolution. LIDAR measurements were also evaluated using a large eddy simulation (LES) of a stable boundary layer that was provided by the National Center for Atmospheric Research. The LIDAR measurement scenario investigated consists of a hub-mounted LIDAR that scans a circle of points upwind of the turbine in order to estimate the wind speed component in the mean wind direction. Different combinations of the preview distance that is located upwind of the rotor and the radius of the scan circle were analyzed. It was found that the dominant source of measurement error for short preview distances is the detection of transverse and vertical wind speeds from the line-of-sight LIDAR measurement. It was discovered in previous studies that, in the absence of wind evolution, the dominant source of error for large preview distances is the spatial averaging caused by the LIDAR's sampling volume. However, by introducing wind evolution, the dominant source of error for large preview distances was found to be the coherence loss caused by evolving turbulence. Different measurement geometries were compared using the bandwidth for which the measurement coherence remained above 0.5 and also the area under the measurement coherence curve. Results showed that, by increasing the intensity of wind evolution, the measurement coherence decreases. Using the coherence bandwidth metric, the optimal preview distance for a fixed-scan radius remained almost constant for low and moderate amounts of wind evolution. For the wind field with the simple wind evolution model introduced, the optimal preview distance for a scan radius of 75% blade span (47.25 meters) was found to be 80 meters. Using the LES wind field, the optimal preview distance was 65 meters. When comparing scan geometries using the area under the coherence curve, results showed that, as the intensity of wind evolution increases, the optimal preview distance decreases.

17 WIND ENERGY↗

Review of Microwave Schottky Beam Diagnostics

A means for non-invasive measurement of transverse and longitudinal characteristics of bunched beams in synchrotrons has been developed based on high sensitivity slotted waveguide pickups. The pickups allow for bandwidths exceeding hundreds of MHz while maintaining good beam sensitivity characteristics. Wide bandwidth is essential to allow bunch-by-bunch measurements by means of a fast gate. The Schottky detector system is installed and successfully commissioned in the Fermilab Tevatron, Recycler and CERN LHC synchrotrons. Measurement capabilities include tune, chromaticity, and momentum spread of single or multiple beam bunches in any combination. With appropriate calibrations, emittance can also be measured by integrating the area under the incoherent tune sidebands.

43 PARTICLE ACCELERATORS↗

SeaStar Unchained: Multiplying the Performance of the Cray SeaStar Network

The Oak Ridge Leadership Computing Facility (OLCF) supports many different systems and many differ- ent interconnects. The only common programming interfaces across these systems are BSD Sockets and MPI. Due to the design assumptions such as implicit buffering leading to extra copies, Sockets performance is almost universally lower than the native interface. Even in the cases that Sockets provides similar bandwidth as the native interface, it suffers from excessive CPU usage. MPI is the de-facto interface for intra- job communication, but is difficult to use between jobs and provides no ability to communicate with service nodes or off- system nodes (e.g. for I/O forwarding). We have developed the Common Communication Interface (CCI), a programming interface that exposes the advances in interconnect hardware, notably Remote Direct Memory Access (RDMA) and operating system (OS) bypass, while imposing as little overhead as possible. This API directly supports inter-job as well as off- system communication. CCI is a lightweight abstraction layer that provides point-to-point messaging and remote memory access. The Cray SeaStar ASIC, with its programmable embedded processor, provides an excellent platform to investigate the properties of various network protocols and programming interfaces. This paper describes our native implementation CCI on the SeaStar platform, and details how we implemented full OS bypass for common operations. We demonstrate a 30% to 50% reduction in latency, more than a six-fold increase in message injection rate, and an almost 7x improvement in bandwidth for small message sizes when compared to the generic Cray Portals implementation.

Dillow, David A↗