Panda: A Compiler Framework for Concurrent CPU $$+$$ + GPU Execution of 3D Stencil Computations on GPU-accelerated Supercomputers

Accelerators like NVIDIA GPUs have changed the landscape of current HPC clusters to a great extent. Massive heterogeneous parallelism offered by these accelerators have led to GPU-Aware MPI libraries that are widely used for writing distributed parallel scientific applications. Compute-oriented collective operations like MPI_Reduce perform computation on data in addition to the usual communication performed by collectives. Historically, these collectives, due to their compute requirements have been implemented on CPU (or Host) only. However, with the advent of GPU technologies it has become important for MPI libraries to provide better design for their GPU (or Device) based versions. In this paper, we tackle the above challenges and provide designs and implementations for most commonly used compute-oriented collectives - MPI_Reduce, MPI_Allreduce, and MPI_Scan - for GPU clusters. We propose extensions to the state-of-the-art algorithms to fully take advantage of the GPU capabilities like GPUDirect RDMA (GDR) and CUDA compute kernel to efficiently perform these operations. With our new designs, we report reduced execution time for all compute-based collectives up to 96 GPUs. Experimental results show an improvement of 50% for small messages and 85% for large messages using MPI_Reduce. For MPI_Allreduce and MPI_Scan, we report more than 40% reduction in time for large messages. Furthermore, analytical models are developed and evaluated to understand and predict the performance of proposed designs for extremely large-scale GPU clusters.

Coarse-Grained reconfigurable architecture (CGRA) is a promising platform for HPC systems in the post-Moore's era. A single-source programming model is essential for practical heterogeneous computing. However, we do not have a canonical programming model and a frontend compiler for it. Existing versatile CGRAs, in respect to their execution model, computational capability, and system structure, magnify the difficulty of orchestrating the compiler techniques. It consequently forces designers of the CGRAs to develop the compiler from scratch, working only for their architectures. Such an approach is outdated, given other successful accelerators like GPU and FPGAs. This paper presents a new CGRA compiler framework in order to reduce development efforts of CG RA applications. OpenMP annotated codes are fed into the proposed compiler, as recent OpenMP support device offloading to the accelerators. This property improves the reusability of the existing source code for HPC workloads. The design of the compiler is inspired by LLVM, which is the most famous compiler framework so that the frontend is built to be architecture-independent. In this work, we demonstrate that the proposed compiler can handle different types of CG RAs without changing the source codes. In addition, we discuss the effect of architecture-independent optimization algorithms. We also provide an open-source implementation of the compiler framework at https://github.com/hal-lab-u-tokyo/CGRAOmp.

Economic optimization and sensitivity analysis of photovoltaic system in residential buildings

PurposeThis research attempts to present a solid transportation problem (STP) mechanism in uncertain and indeterminate contexts, allowing decision makers to select their acceptance, indeterminacy and untruth levels.Design/methodology/approachDue to the lack of reliable information, changeable economic circumstances, uncontrolled factors and especially variable conditions of available resources to adapt to the real situations, the authors are faced with a kind of uncertainty and indeterminacy in constraints and the nature of the parameters of STP. Therefore, an approach based on neutrosophic logic is offered to make it more applicable to real-world circumstances. In this study, the triangular neutrosophic numbers (TNNs) have been utilized to represent demand, transportation capacity, accessibility and cost. Then, the neutrosophic STP was converted into an interval programming problem with the help of the variation degree concept. Then, two simple linear programming models were extracted to obtain the lower and upper bounds of the optimal solution.FindingsThe results reveal that the new model is not complicated but more flexible and more relevant to real-world issues. In addition, it is evident that the suggested algorithm is effective and allows decision makers to specify their acceptance, indeterminacy and falsehood thresholds.Originality/valueUnder the transportation literature, there are several solutions for TP and STP in crisp, fuzzy set (FS) and intuitionistic fuzzy set (IFS) conditions. However, the STP has never been explored in connection with neutrosophic sets to the best of the authors’ knowledge. So, this work tries to fill this gap by coming up with a new way to solve this model using NSs.

Some possibilities in the economic development of Liberia through the forestry sector

This paper is devoted to the management of advance admission requests for obstetric care. Pregnant women in China select one hospital and request admission for both antenatal and postnatal care after nine weeks of pregnancy. Schedulers must make the admission decision instantly based on the availability of the most critical resource, that is, hospital beds for postnatal care. The random delay between admission requests and postnatal care has created a distinct advance admission control problem. To address this issue, we propose a basic model that assumes a unit bed requirement for one day. Each admission generates a unit of revenue and each unit of overcapacity use incurs an overcapacity cost. With the objective of maximizing the expected net revenue, we establish an optimal policy for unlimited requests, that is, an expected arrival time quota (EATQ) policy that accepts a fixed quota of advance admission requests with the same expected date of confinement. We then propose an extended model for general capacity requirements. Using the Poisson approximation, we establish the optimality of the EATQ policy, which is shown to be solvable by a simple linear programming model. We compare the numerical results from the different policies and conduct a sensitivity analysis. The EATQ policy is demonstrated to be the best option in all test instances and notably outperforms the current admission rules used in hospitals, which usually accept admission requests according to some empirical monthly quota of the expected delivery month. The Poisson approximation is shown to be effective for determining the optimal EATQ policy for both stationary and nonstationary arrivals. Summary of Contribution: First, this paper investigates the advance admission control problem for obstetric care. Pregnant women in China choose one hospital and request admission for both antenatal and postnatal care after nine weeks of pregnancy but the most critical resource is hospitalization beds needed for postnatal care. The random delay between admission request and postnatal care makes the problem unique and challenging to solve. It belongs to the scope of computing and operations research. Second, this paper formulates a dynamic programming model, analyzes the structural properties of the optimal control policy, and finally proposes a mathematical programming model to determine the optimal quota. Numerical experiments show the validity of the proposed approach. It covers the research contents of theories on dynamic stochastic control, mathematic programming model, and experiments. Moreover, this paper is motivated by the practical problem (advance admission control) in obstetric units of Shanghai. Using these optimality properties, solution approaches, and numerical results, this paper provides guidance on how to manage advance obstetric admission requests.

As cloud-based computation becomes increasingly important, providing a general computational interface to support datacenter-scale programming has become an imperative research agenda. Many cloud systems use existing virtual machine monitor (VMM) technologies, such as Xen, VMware, and Windows Hypervisor, to multiplex a physical host into multiple virtual hosts and isolate computation on the shared cluster platform. However, traditional multiplexing VMMs do not scale beyond one single physical host, and it alone cannot provide the programming interface and cluster-wide computation that a datacenter system requires. We design a new instruction set architecture, DISA, to unify myriads of compute nodes to form a big virtual machine called DVM, and present programmers the view of a single computer where thousands of tasks run concurrently in a large, unified, and snapshotted memory space. The DVM provides a simple yet scalable programming model and mitigates the scalability bottleneck of traditional distributed shared memory systems. Along with an efficient execution engine, the capacity of a DVM can scale up to support large clusters. We have implemented and tested DVM on three platforms, and our evaluation shows that DVM has excellent performance in terms of execution time and speedup. On one physical host, the system overhead of DVM is comparable to that of traditional VMMs. On 16 physical hosts, the DVM runs 10 times faster than MapReduce/Hadoop and X10. On 256 EC2 instances, DVM shows linear speedup on a parallelizable workload.

We describe a design problem arising in the color printing industry and discuss a number of integer linear programming and constraint programming approaches to its solution. Despite the apparent simplicity of the problem it presents a challenge for both approaches. We present results for three typical cases and show that the constraint programming approach provides better results, although in some cases after considerable handcrafting. We also show that the results obtained by constraint programming can be improved by a simple goal programming model.

This paper presents a security processor to accelerate cryptographic processing in modern security applications. Our security processor is capable of popular cryptographic functions such as RSA, AES, hashing and random number generation, etc. With proposed crypto-DMA controller, data gathering and scattering become flexible for security processing, using a simple descriptor-based programming model. The architecture of the security processor with its core-based platform is scalable and configurable for security variations in performance, cost and power consumption. Different number of data channels and crypto-engines can be used to meet the specifications. In addition, a DFT (design for test) platform is also implemented for the design-test integration. The security processor has been fabricated (using UMC 0.18mum CMOS technology) and measured. The core area is 3.899mm times 2.296mm (525K gates approximately) and the operating clock rate is 66MHz

In MPLS (Multi-Protocol Label Switching) over optical networks, both the optical level and the MPLS level fault recovery can be considered. Generally, a more flexible path arrangement can be realized by the MPLS level recovery, while fast recovery can be achieved by the optical level recovery. When the optical level recovery is adopted, only normal traffic is carried through the working lightpaths and only recovered traffic is carried through the backup lightpaths. In contrast, the working LSPs (Label-Switched Paths) and the backup LSPs corresponding to other working LSPs can be accommodated into an identical lightpath when the MPLS level recovery is adopted. By such sophisticated accommodation of LSPs into the lightpaths, lightpath bandwidth can be utilized efficiently under the condition that the bandwidth utilization is restricted to attain the given objective of transfer quality for the MPLS packets in the normal state and unrestricted in a short time a failure occurs somewhere in the network. This paper proposes a simple mathematical programming model to obtain the optimum arrangement of the working and backup LSPs assuming the MPLS level recovery and a practical LSPs provisioning mode. By comparing the minimized network cost obtained from the optimum arrangement of the working and backup LSPs with the network cost resulting from the optical level recovery, this paper quantitatively evaluates the effectiveness of such bandwidth utilization improvement obtained from the MPLS level recovery and reveals that the MPLS level recovery can actually reduce the network cost due to its flexible arrangement of LSPs on the lightpaths.

Labor Shortages on Small Landholdings in Malawi: Implications for Policy Reforms

This study proposes a new mathematical model that ranks the countries according to their investments. The proposed model compares countries based on specified factors and gives a score for each country. To set out the factors, after investigating on quite a lot of related studies, several macroeconomic factors are gathered, and the most important ones are selected for further investigation. An important advantage of the proposed model is that it can be solved manually by managers, investors, or researchers without a need for any professional optimization solver software. The key factor in this research is developing a mathematical model that ranks any set of countries rather than rating them. At the end of pair comparison, countries are ranked from higher to lower scores from the economical point of view. In addition, the accuracy of the obtained result is validated by comparing them with Moody's rating system (August 2016) using Jaccard similarity index. The results of the proposed are compared with multiple criteria decision‐making techniques as well.

Redundant genotype-to-phenotype mappings are pervasive in evolutionary computation. Such redundancy allows populations to expand in neutral genotypic regions where mutations to a genotype do not alter the phenotypic outcome. Genotype networks have been proposed as a useful framework to characterize the distribution of neutrality among genotypes and phenotypes. In this study, we examine a simple Genetic Programming model that has a finite and compact genotype space by characterizing its genotype networks. We study the topology of individual genotype networks underlying unique phenotypes, investigate the genotypic properties as vertices in genotype networks, and discuss the correlation of these network properties with robustness and evolvability. Using GP simulations of a population, we demonstrate how an evolutionary population diffuses on genotype networks.

Electric power quality is a critical issue for electric utilities and their customers and identification of the power quality disturbances is an important task in power system monitoring and protection. Offline processing of power quality disturbances provides an economic alternative for electric distribution companies, not capable of buying enough number of power quality analyzers for monitoring the disturbances online. Due to the wide frequency range of the disturbances which may happen in a power system, a high sampling rate is necessary for digital processing of the disturbances. Therefore, a large volume of data must be processed for this purpose for each node of an electric distribution network and such a processing has not yet been practical. However, thanks to the rapid developments of digital processors and computer networks, processing big databases is not so hard today. Apache Hadoop is an open-source software framework that allows for the distributed processing of large datasets using simple programming models. In this paper, application of Hadoop distributed computing software for offline processing of power quality disturbances is proposed and it is shown that this application makes such a processing possible and leads to a very cheaper system with widespread usage, compared to the power quality analyzers.

This article presents GRAPE, a parallel GRAPh Engine for graph computations. GRAPE differs from prior systems in its ability to parallelize existing sequential graph algorithms as a whole, without the need for recasting the entire algorithm into a new model. Underlying GRAPE are a simple programming model and a principled approach based on fixpoint computation that starts with partial evaluation and uses an incremental function as the intermediate consequence operator. We show that users can devise existing sequential graph algorithms with minor additions, and GRAPE parallelizes the computation. Under a monotonic condition, the GRAPE parallelization guarantees to converge at correct answers as long as the sequential algorithms are correct. Moreover, we show that algorithms in MapReduce, BSP, and PRAM can be optimally simulated on GRAPE. In addition to the ease of programming, we experimentally verify that GRAPE achieves comparable performance to the state-of-the-art graph systems using real-life and synthetic graphs.