Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Network partitioning”

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 307 records · Page 17

A Latency Tolerant Runtime System for Mass Market Computer Systems

Grappa is a latency tolerant runtime system for mass market computer systems comprised of multiple nodes having x86 multicore processors and infiniband interconnect. Grappa is designed to enable in-memory irregular applications written in the C++ programming language to scale up on multiprocessor systems. Grappa addresses the load balance, communication, and synchronization challenges posed by running applications that do not partition evenly across large systems. Two mechanisms are essential to its success: fast task switching and message aggregation. Fast task switching enables processors to tolerate communication and synchronization delays; aggregation sustains high bandwidth on low injection rate networks.

Source record↗

A Latency Tolerant Runtime System for Mass Market Computer Systems

Grappa is a latency tolerant runtime system for mass market computer systems comprised of multiple nodes having x86 multicore processors and infiniband interconnect. Grappa is designed to enable in-memory irregular applications written in the C++ programming language to scale up on multiprocessor systems. Grappa addresses the load balance, communication, and synchronization challenges posed by running applications that do not partition evenly across large systems. Two mechanisms are essential to its success: fast task switching and message aggregation. Fast task switching enables processors to tolerate communication and synchronization delays; aggregation sustains high bandwidth on low injection rate networks.

Source record↗

Coupling of organic and inorganic aerosol systems and the effect on gas–particle partitioning in the southeastern US

Abstract. Several models were used to describe the partitioning of ammonia, water, and organic compounds between the gas and particle phases for conditions in the southeastern US during summer 2013. Existing equilibrium models and frameworks were found to be sufficient, although additional improvements in terms of estimating pure-species vapor pressures are needed. Thermodynamic model predictions were consistent, to first order, with a molar ratio of ammonium to sulfate of approximately 1.6 to 1.8 (ratio of ammonium to 2 × sulfate, R N/2S ≈ 0.8 to 0.9) with approximately 70 % of total ammonia and ammonium (NH x ) in the particle. Southeastern Aerosol Research and Characterization Network (SEARCH) gas and aerosol and Southern Oxidant and Aerosol Study (SOAS) Monitor for AeRosols and Gases in Ambient air (MARGA) aerosol measurements were consistent with these conditions. CMAQv5.2 regional chemical transport model predictions did not reflect these conditions due to a factor of 3 overestimate of the nonvolatile cations. In addition, gas-phase ammonia was overestimated in the CMAQ model leading to an even lower fraction of total ammonia in the particle. Chemical Speciation Network (CSN) and aerosol mass spectrometer (AMS) measurements indicated less ammonium per sulfate than SEARCH and MARGA measurements and were inconsistent with thermodynamic model predictions. Organic compounds were predicted to be present to some extent in the same phase as inorganic constituents, modifying their activity and resulting in a decrease in [H + ] air (H + in µg m -3 air), increase in ammonia partitioning to the gas phase, and increase in pH compared to complete organic vs. inorganic liquid–liquid phase separation. In addition, accounting for nonideal mixing modified the pH such that a fully interactive inorganic–organic system had a pH roughly 0.7 units higher than predicted using traditional methods (pH = 1.5 vs. 0.7). Particle-phase interactions of organic and inorganic compounds were found to increase partitioning towards the particle phase (vs. gas phase) for highly oxygenated (O : C ≥ 0.6) compounds including several isoprene-derived tracers as well as levoglucosan but decrease particle-phase partitioning for low O : C, monoterpene-derived species.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Navier-Stokes Computations on Commodity Computers

In this paper we discuss and demonstrate the feasibility of solving high-fidelity, nonlinear computational fluid dynamics (CFD) problems of practical interest on commodity machines, namely Pentium Pro PC's. Such calculations have now become possible due to the progress in computational power and memory of the off-the-shelf commodity computers, along with the growth in bandwidth and communication speeds of networks. A widely used CFD code known as TLNS3D, which was developed originally on large shared memory computers was selected for this effort. This code has recently been ported to massively parallel processor (MPP) type machines, where natural partitioning along grid blocks is adopted in which one or more blocks are distributed to each of the available processors. In this paper, a similar approach is adapted to port this code to a cluster of Pentium Pro computers. The message passing among the processors is accomplished through the use of standard message passing interface (MPI) libraries. Scaling studies indicate fairly high level of parallelism on such clusters of commodity machines, thus making solutions to Navier-Stokes equations for practical problems more affordable.

Vatsa, Veer N.↗

Identifying Differences in Abiotic Stress Gene Networks between Lowland and Upland Ecotypes of Switchgrass (DE-SC0008338)

Switchgrass (Panicum virgatum) is a warm-season C4 grass that is a target lignocellulosic biofuel species for use in the United States due to its local adaption capabilities and high biomass accumulation. Two ecotypes of switchgrass have been described. Members of the lowland ecotype are taller, have narrower leaf blades and generate more biomass compared to individuals from the upland ecotype. Additionally, lowland plants are generally found in the southern United States while upland switchgrass is more typically present in the northern United States. These differences are important as it is envisioned that switchgrass for biofuel production will typically be grown on marginal lands in the northern United States to supplement and diversify farmers' traditional crop incomes. While lowland switchgrass is more productive, it has poor winter survivability in northern latitudes where upland switchgrass is expected to be grown for biofuel use. Abiotic stresses likely to be encountered by switchgrass include drought and salinity. Despite initially being described as preferring wetter environments, members of the lowland ecotype have been characterized as being more drought tolerant than plants of the upland ecotype. Nonetheless, direct trials have indicated that variation for drought tolerance exists in both ecotypes, but prior to this project, only a relatively small number of switchgrass lines had been tested for drought responses. Similarly, switchgrass cultivars have not been widely tested for salt tolerance, but a few studies have shown that even mild salt stress can inhibit growth. The effects of drought and salt stress on plant growth are complex. Both drought and salinity affect the osmotic potential of plant cells and negatively affect plant growth due to reduced water potential and reduced photosynthesis that results from lower stomatal conductance of CO 2 . Plants respond to drought and salt stress by activating genes that directly attempt to reduce the stress (e.g., transmembrane pumps that partition Na + ) and mitigate the effects of the stress (e.g., synthesis of osmoprotectant metabolites and stress-related signaling compounds). Prior to the start of this project, no gene expression analysis had been performed on switchgrass under conditions of drought or salt stress, and therefore, relevant gene networks responding to drought and salt stress were unknown in switchgrass. In this project, we performed drought, salt and alkali-salt screens on 49 switchgrass cultivars (Liu et al 2014; Liu et al 2015; Hu et al 2015; Kim et al 2016). These experiments demonstrated that a wide range of variation exists within switchgrass for drought, salt and alkali-salt tolerance and that, while the lowland ecotype of switchgrass is often considered more tolerant of abiotic stresses, there are some upland switchgrass lines that are also very tolerant of drought, salt and alkali-salt stress. We also conducted drought and salt time course experiments with Alamo and Dacotah. We have identified modules of coexpressed genes that differentiate Alamo and Dacotah drought responses. We are continuing to analyze these results and plan to submit manuscripts describing this work in early 2017. In an effort to show how drought- and salt-related gene modules could be dissected, we generated transgenic switchgrass overexpressing either PvGTγ-1 or ZmDREB2. Increased expression of PvGTγ-1 does confer increased salt tolerance, and we were able to identify genes that are induced and suppressed by PvGTγ-1. Overexpression of ZmDREB2 increases drought tolerance in switchgrass. Analysis of the PvGTγ-1 and ZmDREB2 overexpression work is ongoing, and we plan to prepare manuscripts about these experiments for submission in early 2017.

59 BASIC BIOLOGICAL SCIENCES↗

Uncertainty quantification for nuclear forensics with population analyses

Although neural networks offer cutting-edge predictive power, their deployment in high-consequence nuclear forensic applications is limited, partly because of their black-box nature. Incorporating robust uncertainty quantification methods into the predictive frameworks of neural networks is progress towards their future deployment in such scenarios. This work integrates uncertainty quantification into neural networks for nuclear reactor core-average burnup estimation from simulated environmental samples. We test two regimes (homogeneous and heterogeneous events) on DeepSets and Set Transformer architectures, we find both quantify predictive uncertainty effectively, but Set Transformer excels in partitioning latent events, offering superior predictive power and more informative uncertainty estimates.

Hatton, Conner [ORNL] (ORCID:0009000804970959)↗

Explicit spatial scattering for load balancing in conservatively synchronized parallel discrete-event simulations

We re-examine the problem of load balancing in conservatively synchronized parallel, discrete-event simulations executed on high-performance computing clusters, focusing on simulations where computational and messaging load tend to be spatially clustered. Such domains are frequently characterized by the presence of geographic 'hot-spots' - regions that generate significantly more simulation events than others. Examples of such domains include simulation of urban regions, transportation networks and networks where interaction between entities is often constrained by physical proximity. Noting that in conservatively synchronized parallel simulations, the speed of execution of the simulation is determined by the slowest (i.e most heavily loaded) simulation process, we study different partitioning strategies in achieving equitable processor-load distribution in domains with spatially clustered load. In particular, we study the effectiveness of partitioning via spatial scattering to achieve optimal load balance. In this partitioning technique, nearby entities are explicitly assigned to different processors, thereby scattering the load across the cluster. This is motivated by two observations, namely, (i) since load is spatially clustered, spatial scattering should, intuitively, spread the load across the compute cluster, and (ii) in parallel simulations, equitable distribution of CPU load is a greater determinant of execution speed than message passing overhead. Through large-scale simulation experiments - both of abstracted and real simulation models - we observe that scatter partitioning, even with its greatly increased messaging overhead, significantly outperforms more conventional spatial partitioning techniques that seek to reduce messaging overhead. Further, even if hot-spots change over the course of the simulation, if the underlying feature of spatial clustering is retained, load continues to be balanced with spatial scattering leading us to the observation that spatial scattering can often obviate the need for dynamic load balancing.

97 MATHEMATICS AND COMPUTING↗

Parallel and Distributed Computational Fluid Dynamics: Experimental Results and Challenges

This paper describes several results of parallel and distributed computing using a large scale production flow solver program. A coarse grained parallelization based on clustering of discretization grids combined with partitioning of large grids for load balancing is presented. An assessment is given of its performance on distributed and distributed-shared memory platforms using large scale scientific problems. An experiment with this solver, adapted to a Wide Area Network execution environment is presented. We also give a comparative performance assessment of computation and communication times on both the tightly and loosely-coupled machines.

Djomehri, Mohammad Jahed↗

Overlapping clusters for distributed computation.

Scalable, distributed algorithms must address communication problems. We investigate overlapping clusters, or vertex partitions that intersect, for graph computations. This setup stores more of the graph than required but then affords the ease of implementation of vertex partitioned algorithms. Our hope is that this technique allows us to reduce communication in a computation on a distributed graph. The motivation above draws on recent work in communication avoiding algorithms. Mohiyuddin et al. (SC09) design a matrix-powers kernel that gives rise to an overlapping partition. Fritzsche et al. (CSC2009) develop an overlapping clustering for a Schwarz method. Both techniques extend an initial partitioning with overlap. Our procedure generates overlap directly. Indeed, Schwarz methods are commonly used to capitalize on overlap. Elsewhere, overlapping communities (Ahn et al, Nature 2009; Mishra et al. WAW2007) are now a popular model of structure in social networks. These have long been studied in statistics (Cole and Wishart, CompJ 1970). We present two types of results: (i) an estimated swapping probability {rho}{infinity}; and (ii) the communication volume of a parallel PageRank solution (link-following {alpha} = 0.85) using an additive Schwarz method. The volume ratio is the amount of extra storage for the overlap (2 means we store the graph twice). Below, as the ratio increases, the swapping probability and PageRank communication volume decreases.

99 GENERAL AND MISCELLANEOUS↗

Evidence of Technetium and Iodine from a Sodalite-Bearing Ceramic Waste Form

Current plans for nuclear waste vitrification at the Hanford Tank Waste Treatment and Immobilization Plant (WTP) lack the capacity to treat all of the low activity waste (LAW) that is not encapsulated in the vitrified product. Several technologies are being considered to treat the excess LAW. One such technology is Fluidized Bed Steam Reforming (FBSR). The FBSR process results in a granular product composed of feldspathoid mineral phases that immobilize the major components in the LAW as well as other contaminants of concern (COCs), with Tc and I expected to be present in sodalite cages formed during the process. In order to meet compressive strength requirements at the Hanford Integrated Disposal Facility (IDF), the granular product may be encapsulated in a monolith. To demonstrate the ability of the technology to serve the mission of managing excess LAW, Single Pass Flow-Through (SPFT) tests have been performed on non-radioactive granular materials and granular materials encapsulated in a geopolymer binder produced at the engineering- and bench-scale as well as a granular product produced at the bench scale with actual Hanford tank waste. SPFT tests were conducted at 40 °C for durations up to 2 months with a flow-through solution buffered at pH 9. The forward reaction rate of the non-radioactive mineral product dissolution based on Si release for the granular product was measured to be (6.2 ± 2.1) × 10-4 g/m2d for the engineering-scale product and (13 ± 4.9) × 10-4 g/m2d for the bench-scale product. The resulting non-radioactive monoliths showed forward reaction rates based on Si release of (3.4 ± 1.1) × 10-4 g/m2d for the engineering-scale material and (4.2 ± 1.5) × 10-4 g/m2d for the bench-scale material demonstrating that encapsulation of the FBSR granular product in a monolith does not significantly alter the performance of the material. Finally, an FBSR granular product created at the bench scale using actual Hanford LAW gave similar release values and also demonstrated that Re and Tc show similar release behavior with a dilute-condition release rate of (1.1 ± 0.5) × 10-4 g/m2d for Tc and (1.6 ± 0.6) × 10-4 g/m2d for Re. We believe that this is the first demonstration to show that Re and I have similar release behavior and that similar behavior of Re and Tc also occurs. These results indicate that the FBSR mineral product may be a suitable supplementary immobilization technique at the Hanford site and they also suggest that both Tc and Re occupy similar sites in the sodalite cage structure.

fluidized bed steam reforming↗

Performance modeling and measurement of real-time multiprocessors with time-shared buses

A closed queueing network model is constructed to address workload effects on computer performance for a highly reliable unibus multiprocessor used in real-time control. The queueing model consists of multiserver nodes and a nonpreemptive priority queue. Use of this model requires partitioning the workload into task classes. The time average steady-state solution of the queueing model directly produces useful results that are necessary in performance evaluation. The model is experimentally justified with the Fault-Tolerant Multiprocessor (FTMP) located at the NASA AIRLAB. Extensive experiments are performed on FTMP with a synthetic workload generator (SWG) to directly measure performance parameters, such as processor idle time, system bus contention, and task processing times. These measurements determine values for parameters in the queueing model. Experimental and analytic results are then compared.

Woodbury, Michael H.↗

Designing Energy Efficient Communication Runtime Systems: A View from PGAS Models

As the march to the Exascale computing gains momentum, energy consumption of supercomputers has emerged to be the critical roadblock. While architectural innovations are imperative in achieving computing of this scale, it is largely dependent on the systems software to leverage the architectural innovations. Parallel applications in many computationally intensive domains have been designed to leverage these supercomputers, with legacy two-sided communication semantics using Message Passing Interface. At the same time, Partitioned Global Address Space Models are being designed which provide global address space abstractions and one-sided communication for exploiting data locality and communication optimizations. PGAS models rely on one-sided communication runtime systems for leveraging high-speed networks to achieve best possible performance. In this paper, we present a design for Power Aware One-Sided Communication Library - PASCoL. The proposed design detects communication slack, leverages Dynamic Voltage and Frequency Scaling (DVFS) and Interrupt driven execution to exploit the detected slack for energy efficiency. We implement our design and evaluate it using synthetic benchmarks for one-sided communication primitives, Put, Get and Accumulate and uniformly non-contiguous data transfers. Our performance evaluation indicates that we can achieve significant reduction in energy consumption without performance loss on multiple one-sided communication primitives. The achieved results are close to the theoretical peak available with the experimental test bed.

Vishnu, Abhinav↗

Efficient On-demand Connection Management Mechanisms with PGAS Models on InfiniBand

In the last decade or so, clusters have observed a tremendous rise in popularity due to the excellent price to performance ratio. A variety of Interconnects have been proposed during this period, with InfiniBand leading the way due to its high performance and open standard. At the same time, multiple programming models have emerged in order to meet the requirements of various applications and their programming models. To support requirements of multiple programming models, InfiniBand provides multiple transport semantics, ranging from unreliable connectionless to reliable connected characteristics. Among them, the reliable connection (RC) semantics is being widely due to its high performance and support for novel features like Remote Direct Memory Acesss (RDMA), hardware atomics and Network Fault Tolerance. However, the pairwise connection oriented nature of the RC transport semantics limits its scalability and usage at the increasing processor counts. In this paper, we design and implement on-demand connection management approaches in the context of Partitioned Global Address Space (PGAS) programming models, which provided shared memory abstraction and one-sided communication semantics, leading to the development of multiple languages (UPC, X10, Chapel) and libraries (Global Arrays, MPI-RMA). Using Global Arrays as the research vehicle, we implement this approach with Aggregate Remote Memory Copy Interface (ARMCI), the runtime system of Global Arrays. We evaluate our approach, ARMCI-On Demand Connection Management (ARMCI-ODCM) using various microbenchmarks and benchmarks (LU Factorization, RandomAccess and Lennard Jones simulation) and application (Subsurface transport over multiple phases (STOMP)). With the performance evaluation for up to 4096 processors, we are able to have a multi-fold reduction in connection memory with a negligible degradation in performance. Using STOMP at 4096 processors, reduces the overall connection memory by 66 times with no performance degradation. To the best of our knowledge, this is the first design, implementation and evaluation of on-demand connection management with InfiniBand using PGAS models.

54 ENVIRONMENTAL SCIENCES↗

Altered Actin Centripetal Retrograde Flow in Physically Restricted Immunological Synapses

Antigen recognition by T cells involves large scale spatial reorganization of numerous receptor, adhesion, and costimulatory proteins within the T cell-antigen presenting cell (APC) junction. The resulting patterns can be distinctive, and are collectively known as the immunological synapse. Dynamical assembly of cytoskeletal network is believed to play an important role in driving these assembly processes. In one experimental strategy, the APC is replaced with a synthetic supported membrane. An advantage of this configuration is that solid structures patterned onto the underlying substrate can guide immunological synapse assembly into altered patterns. Here, we use mobile anti-CD3e on the spatial-partitioned supported bilayer to ligate and trigger T cell receptor (TCR) in live Jurkat T cells. Simultaneous tracking of both TCR clusters and GFP-actin speckles reveals their dynamic association and individual flow patterns. Actin retrograde flow directs the inward transport of TCR clusters. Flow-based particle tracking algorithms allow us to investigate the velocity distribution of actin flow field across the whole synapse, and centripetal velocity of actin flow decreases as it moves toward the center of synapse. Localized actin flow analysis reveals that, while there is no influence on actin motion from substrate patterns directly, velocity differences of actin are observed over physically trapped TCR clusters. Actin flow regains its velocity immediately after passing through confined TCR clusters. These observations are consistent with a dynamic and dissipative coupling between TCR clusters and viscoelastic actin network.

59 BASIC BIOLOGICAL SCIENCES↗

The polarity protein Par6 is coupled to the microtubule network during molluscan early embryogenesis

Research highlights: {yields} The cDNAs encoding Par6 and aPKC homologues were cloned from the snail Lymnaea stagnalis. {yields} L. stagnalis Par6 directly interacts with tubulin and microtubules and localizes to the microtubule cytoskeleton during the early embryogenesis. {yields} Identical sequence and localization of LsPar6 for the dextral and the sinistral snails exclude the possibility of the gene being the primary determinant of body handedness. -- Abstract: Cell polarity, which directs the orientation of asymmetric cell division and segregation of fate determinants, is a fundamental feature of development and differentiation. Regulators of polarity have been extensively studied, and the critical importance of the Par (partitioning-defective) complex as the polarity machinery is now recognized in a wide range of eukaryotic systems. The Par polarity module is evolutionarily conserved, but its mechanism and cooperating factors vary among different systems. Here we describe the cloning and characterization of a pond snail Lymnaea stagnalis homologue of partitioning-defective 6 (Lspar6). The protein product LsPar6 shows high affinity for microtubules and localizes to the mitotic apparatus during embryonic cell division. In vitro assays revealed direct binding of LsPar6 to tubulin and microtubules, which is the first evidence of the direct interaction between the two proteins. The interaction is mediated by two distinct regions of LsPar6 both located in the N-terminal half. Atypical PKC, a functional partner of Par6, was also found to localize to the mitotic spindle. These results suggest that the L. stagnalis Par complex employs the microtubule network in cell polarity processes during the early embryogenesis. Identical sequence and localization of LsPar6 for the dextral and the sinistral snails exclude the possibility of the gene being the primary determinant of handedness.

60 APPLIED LIFE SCIENCES↗

Coordinated platooning with multiple speeds

In a platoon, vehicles travel one after another with small intervehicle distances; trailing vehicles in a platoon save fuel because they experience less aerodynamic drag. This work presents a coordinated platooning model with multiple speed options that integrates scheduling, routing, speed selection, and platoon formation/dissolution in a mixed-integer linear program that minimizes the total fuel consumed by a set of vehicles while traveling between their respective origins and destinations. The performance of this model is numerically tested on a grid network and the Chicago-area highway network. We find that the fuel-savings factor of a multivehicle system significantly depends on the time each vehicle is allowed to stay in the network; this time affects vehicles’ available speed choices, possible routes, and the amount of time for coordinating platoon formation. For problem instances with a large number of vehicles, we propose and test a heuristic decomposed approach that applies a clustering algorithm to partition the set of vehicles and then routes each group separately. When the set of vehicles is large and the available computational time is small, the decomposed approach finds significantly better solutions than does the full model.

33 ADVANCED PROPULSION SYSTEMS↗

Self-Assembled Bolaamphiphile-Based Organic Nanotubes as Efficient Cu(II) Ion Adsorbents

Self-assembled organic nanotubes (ONTs) have been actively examined for various applications such as chemical separations and catalysis owing to their well-defined tubular nanostructures with distinct chemical environments at the wall and internal/external surfaces. Adsorption of heavy metal ions onto ONTs plays an essential role in many of these applications, but it has rarely been assessed quantitatively. Herein, we investigated interactions between Cu 2+ and single-/quadruple-wall bolaamphiphile-based ONTs having inner carboxyl groups with different inner diameters, COOH-ONT 10nm and COOH-ONT 20nm . We first examined the effects of Cu 2+ on their nanotubular structures using SAXS, STEM, and AFM. COOH-ONT 10nm was stable in aqueous Cu 2+ solution in contrast to COOH-ONT 20nm owing to the presence of polyglycine-II-type hydrogen bonding networks within its wall. Subsequently, we studied the Cu 2+ adsorption behavior of COOH-ONT 10nm by monitoring the concentration of unbound Cu 2+ using linear sweep anodic stripping voltammetry. The Cu 2+ adsorption was quick, attributable to efficient Cu 2+ partitioning through the open ends of the ONT, followed by fast Cu 2+ diffusion in the uniform, relatively large nanochannel. More importantly, the Cu 2+ adsorption capacity and affinity of COOH-ONT 10nm were measured at different pH using the Langmuir adsorption model. The adsorption capacity was similar at the pH range examined, showing the participation of approximately 25% of the inner carboxyl groups in the adsorption. The adsorption affinity increased with pH, indicating the essential role of the deprotonated carboxyl groups in the Cu 2+ adsorption. Most interestingly, the Langmuir adsorption constant was significantly higher than those of previously reported synthetic adsorbents and planar monolayer based on carboxyl binding sites. The high Cu 2+ affinity of the ONT was attributable to the highly dense binding sites on the well-defined nanoscale concave structure of the inner channel. Furthermore, these results provide a valuable guideline to designing self-assembled nanomaterials for efficient chemical separations, detection, and catalysis.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

An Initial Implementation of Libfabric Conduit for OpenSHMEM-X

As a representative of Partitioned Global Address Space models, OpenSHMEM provides a variety of functionalities including one-sided communication, atomic operations, and collective routines. The communication layer of OpenSHMEM-X plays a crucial role for its functionalities. OFI Libfabric is an open-source network library that supports portable low-latency interfaces from different fabric providers while minimizing the semantic gap across API endpoints. In this paper, we present the design and implementation of OpenSHMEM-X communication conduit using Libfabric. This Libfabric conduit is designed to support a broad range of network providers while achieving excellent network performance and scalability. We have performed an extensive set of experiments to validate the performance of our implementation, and compared with the Sandia OpenSHMEM implementation. Our results show that the Libfabric conduit improves the communication bandwidth on the socket provider by up to 42% and 11%, compared to an alternative OpenSHMEM implementation for put and get operations, respectively. In addition, our implementation of atomic operations has achieved similar latency to that of the Sandia implementation.

Gorentla Venkata, Manjunath↗