MST: Topology-Aware Message Aggregation for Exascale Graph Processing of Traversal-Centric Algorithms

Large-scale research infrastructures (LSRIs) are widely acknowledged as a crucial instrument for venturing into the uncharted territories of science and technology, as well as contributing to the well-being of society. However, only a limited number of literature have scrutinized the impact of LSRIs, founded upon a causal inference framework. Moreover, the function of LSRIs in the advancement of innovation at the regional level remains inadequately identified. Drawing on the resource-based view, this study develops a conceptual framework that links the scientific effect of LSRIs to innovation resources in order to assess their impact on knowledge innovation (KI). Taking China’s National Supercomputing Center (NSC) as a case, three major mechanism hypotheses are proposed for the impact of NSC on KI, including basic effect, network effect, and technology effect. Using panel data from 283 cities in China from 2000 to 2020, we employ a spatial difference-in-differences estimation model to examine the impact of NSC on KI. The research finds that: (1) The construction of NSC stimulates KI in local and surrounding areas. (2) The main mechanisms by which NSC promotes KI include the increase in fiscal investment and talents in science, the improvement of digital infrastructure, as well as the enhancement of urban network centrality and innovation efficiency. (3) Geographical proximity, cooperation proximity, and digitization proximity constitute the main channels of policy spillover. (4) NSC has not shown significant promotion of regional innovation convergence, and its radiation influence needs further improvement. (5) The knowledge innovation effects of NSCs manifest heterogeneity based on the distinct knowledge orientation and innovation environment, with this impact being notably pronounced in application innovation-oriented cities such as Shenzhen. The results of this study reveal the positive yet limited impact of NSC on KI and provide a reference for other economies in the areas of LSRIs, digital infrastructure, and the formulation of place-based innovation policy.

Graph algorithms such as graph traversal have been gaining ever-increasing importance in the era of big data. However, graph processing on traditional architectures issues many random and irregular memory accesses, leading to a huge number of data movements and the consumption of very large amounts of energy. To minimize the waste of memory bandwidth, we investigate utilizing processing-in-memory (PIM), combined with non-volatile metal-oxide resistive random access memory (ReRAM), to improve both computation and I/O performance. We propose a new ReRAM-based processing-in-memory architecture called RPBFS, in which graph data can be persistently stored and processed in place. We study the problem of graph traversal, and we design an efficient graph traversal algorithm in RPBFS. Benefiting from low data movement overhead and high bank-level parallel computation, RPBFS shows a significant performance improvement compared with both the CPU-based and the GPU-based BFS implementations. On a suite of real-world graphs, our architecture yields a speedup in graph traversal performance of up to 33.8×, and achieves a reduction in energy over conventional systems of up to 142.8×.

This paper describes the design and performance evaluation of the mechanisms for latency tolerance in the remote memory access communication on clusters equipped with high-performance networks such as Myrinet. It discusses strategies that bridge the gap between user-level requirements and network-specific communication interfaces while attempting to increase opportunities for latency hiding. Mechanisms for overlapping communication with computation and coalescing small messages (trading latency for bandwidth) are explored. The effectiveness of these techniques is evaluated using microbenchmarks and application kernels including the NAS parallel benchmark suite. The microbenchmark results showed a much better degree of overlap for non-blocking operations in ARMCI when compared with MPI. Application results showed up to 30 45% improvement over MPI on using non-blocking operations. The aggregation of small messages yielded performance improvement of up to 78% over non-aggregated communication.

This paper investigates the power, energy, and performance characteristics of large-scale graph processing on hybrid (i.e., CPU and GPU) single-node systems. Graph processing can be accelerated on hybrid systems by properly mapping the graph-layout to processing units, such that the algorithmic tasks exercise each of the units where they perform best. However, the GPUs have much higher Thermal Design Power (TDP), thus their impact on the overall energy consumption is unclear. Our evaluation using large real-world graphs and synthetic graphs as large as 1 billion vertices and 16 billion edges shows that a hybrid system is efficient in terms of both time-to-solution and energy.

MotivationSynthetic lethality (SL) is a form of genetic interaction that can selectively kill cancer cells without damaging normal cells. Exploiting this mechanism is gaining popularity in the field of targeted cancer therapy and anticancer drug development. Due to the limitations of identifying SL interactions from laboratory experiments, an increasing number of research groups are devising computational prediction methods to guide the discovery of potential SL pairs. Although existing methods have attempted to capture the underlying mechanisms of SL interactions, methods that have a deeper understanding of and attempt to explain SL mechanisms still need to be developed.ResultsIn this work, we propose a novel SL prediction method, SLGNN. This method is based on the following assumption: SL interactions are caused by different molecular events or biological processes, which we define as SL-related factors that lead to SL interactions. SLGNN, apart from identifying SL interaction pairs, also models the preferences of genes for different SL-related factors, making the results more interpretable for biologists and clinicians. SLGNN consists of three steps: first, we model the combinations of relationships in the gene-related knowledge graph as the SL-related factors. Next, we derive initial embeddings of genes through an explicit message aggregation process of the knowledge graph. Finally, we derive the final gene embeddings through an SL graph, constructed using known SL gene pairs, utilizing factor-based message aggregation. At this stage, a supervised end-to-end training model is used for SL interaction prediction. Based on experimental results, the proposed SLGNN model outperforms all current state-of-the-art SL prediction methods and provides better interpretability.Availability and implementationSLGNN is freely available at https://github.com/zy972014452/SLGNN.

The paper describes graph algorithms performance when using different types of data structures. To achieve that, we developed a multi-level graph processing system, which allows to create graph applications independently of any implementation details such as graph data structure or underlying computational architecture. We measure the performance of breadth-first search, max flow and random graph building algorithms when using compressed sparse row and adjacency matrix data structures. Experiments reveal different graph processing rates for different data structures, which indicates the need of using specific data structures for specific algorithms to achieve highest performance.

Many big data analytics applications explore a set of related entities, which are naturally modeled as graph. However, graph processing is notorious for its performance challenges due to random data access patterns, especially for large data volumes. Solving these challenges is critical to the performance of industry-scale applications. In contrast to most prior works, which focus on accelerating a single graph processing task, in industrial practice we consider multiple graph processing tasks running concurrently, such as a group of queries issued simultaneously to the same graph. In this paper, we present an edge-set based graph traversal framework called C-Graph (i.e. Concurrent Graph), running on a distributed infrastructure, that achieves both high concurrency and efficiency for k-hop reachability queries. The proposed framework maintains global vertex states to facilitate graph traversals, and supports both synchronous and asynchronous communication. In this study, we decompose a set of graph processing tasks into local traversals and analyze their performance on C-Graph. More specifically, we optimize the organization of the physical edge-set and explore the shared subgraphs. We experimentally show that our proposed framework outperforms several baseline methods.

The purpose of this paper is to develop a specialized language for processing graph data and its software implementation. The proposed solution ensures versatility, usability, and efficiency, enabling the execution of both basic graph operations and more complex procedures. The tool supports classical graph algorithms, including shortest path search, graph traversal, and minimum spanning tree construction, as well as applications in modeling and analyzing message transition processes and processing graph-based representations of textual data. A common drawback of many existing graph analysis tools—often implemented as libraries of general-purpose programming languages—is their limited usability. This limitation arises from the fact that the description of graph analysis procedures relies on data structure operations defined in terms of these general-purpose languages, which complicates perception and reduces the clarity of the mathematical methods being implemented. Developing a specialized domain-specific language based on high-level abstractions can address these shortcomings. Such a language will provide a formalized description of methods for analyzing and processing graph models, improving their comprehensibility and accessibility to users. Its software implementation will deliver ready-to-use solutions for executing graph analysis methods. Composing such methods will facilitate solving a wide range of tasks, including the analysis of natural language messages and the study of information publication and dissemination processes in online environments.

Nowadays, Infrastructure-as-a-Service (IaaS) cloud become popular to provide more and more compute powers. The importance of resources management in the private cloud clusters raise correspondingly, as it's the significate guarantee to the service quality. When IaaS providers provide user large scale of virtual machine instances, users can't stand long for the preparation phase of the instances initialization, especially when the scale is really large. In this paper, we present a reality check of IaaS cloud large file dissemination. We deploy the experiments in National Super Computing Center in Tianjin (NSCC-TJ), China, which use multiple snowball tree algorithm to dissemination virtual machine images, software packages, software or system kernel updates in large quantity. We optimize and accelerate the algorithm to fit our system and environment. As a comparison, it speed up the image dissemination progress to 1 hour from 30 hours using Eucalyptus.

In today's knowledge-, service-, and cloud-based economy, businesses accumulate massive amounts of data from a variety of sources. In order to understand businesses one may need to perform considerable analytics over large hybrid collections of heterogeneous and partially unstructured data that is captured related to the process execution. This data, usually modeled as graphs, increasingly come to show all the typical properties of big data: wide physical distribution, diversity of formats, non-standard data models, independently-managed and heterogeneous semantics. We use the term big process graph to refer to such large hybrid collections of heterogeneous and partially unstructured process related execution data. Online analytical processing (OLAP) of big process graph is challenging as the extension of existing OLAP techniques to analysis of graphs is not straightforward. Moreover, process data analysis methods should be capable of processing and querying large amount of data effectively and efficiently, and therefore have to be able to scale well with the infrastructure's scale. While traditional analytics solutions (relational DBs, data warehouses and OLAP), do a great job in collecting data and providing answers on known questions, key business insights remain hidden in the interactions among objects: it will be hard to discover concept hierarchies for entities based on both data objects and their interactions in process graphs. In this paper, we introduce a framework and a set of methods to support scalable graph-based OLAP analytics over process execution data. The goal is to facilitate the analytics over big process graph through summarizing the process graph and providing multiple views at different granularity. To achieve this goal, we present a model for process OLAP (P-OLAP) and define OLAP specific abstractions in process context such as process cubes, dimensions, and cells. We present a MapReduce-based graph processing engine, to support big data analytics over process graphs. We have implemented the P-OLAP framework and integrated it into our existing process data analytics platform, ProcessAtlas, which introduces a scalable architecture for querying, exploration and analysis of large process data. We report on experiments performed on both synthetic and real-world datasets that show the viability and efficiency of the approach.

We introduce Episodic Memory QA, the task of answering personal user questions grounded on memory graph (MG), where episodic memories and related entity nodes are connected via relational edges. We create a new benchmark dataset first by generating synthetic memory graphs with simulated attributes, and by composing 100K QA pairs for the generated MG with bootstrapped scripts. To address the unique challenges for the proposed task, we propose Memory Graph Networks (MGN), a novel extension of memory networks to enable dynamic expansion of memory slots through graph traversals, thus able to answer queries in which contexts from multiple linked episodes and external knowledge are required. We then propose the Episodic Memory QA Net with multiple module networks to effectively handle various question types. Empirical results show improvement over the QA baselines in top-k answer prediction accuracy in the proposed task. The proposed model also generates a graph walk path and attention vectors for each predicted answer, providing a natural way to explain its QA reasoning.

This article presents electricity demand forecasting models for industrial and residential facilities, developed using ensemble machine learning strategies. Short term electricity demand forecasting is beneficial for both consumers and suppliers, as it allows improving energy efficiency policies and the rational use of resources. Computational intelligence models are developed for day-ahead electricity demand forecasting. An ensemble strategy is applied to build the day-ahead forecasting model based on several one-hour models. Three steps of data preprocessing are carried out, including treating missing values, removing outliers, and standardization. Feature extraction is performed to reduce overfitting, reducing the training time and improving the accuracy. The best model is optimized using grid search strategies on hyperparameter space. Then, an ensemble of 24 instances is generated to build the complete day-ahead forecasting model. Considering the computational complexity of the applied techniques, they are developed and evaluated on the National Supercomputing Center (Cluster-UY), Uruguay. Three different real data sets are used for evaluation: an industrial park in Burgos (Spain), the total electricity demand for Uruguay, and demand from a distribution substation in Montevideo (Uruguay). Standard performance metrics are applied to evaluate the proposed models. The main results indicate that the best day ahead model based on ExtraTreesRegressor has a mean absolute percentage error of 2:55% on industrial data, 5:17% on total consumption data and 9:09% on substation data.

In this paper, we present an instructional HPC-based platform, ModStore, for Singapore students to learn data science through National Science Experiment (NSE) big data challenge. Before the challenge, Singapore students, ranging from secondary school, high school to junior college, polytechnics and ITE, carry specially designed sensors to collect data on their daily travel and the surrounding environment during the experiment. However, most of the students do not have any programming background to do data science for this big data challenge. To address this issue, we develop ModStore, powered by National Supercomputing Centre, Singapore, which serves as a powerful tool to enable students to learn about the Internet of Things and Big Data in a drag-and-drop manner by building the workflow, analyzing the information, creating graphs, interpreting visualizations, and comparing trends. 180 students were using ModStore to participate this challenge from October to December, 2016. It has been touted as a promising instructional platform for learning Mathematics, statistics and data science.

This article describes the effort to join two different approaches for the development of the proactive network management. The first approach was developed in the Laboratory of Network and Management, at Federal University of Santa Catarina, which uses remote monitoring and simulation tools. The second one, was developed in the National Supercomputing Center, which handles the problem with artificial intelligence and SunNet Manager's agents usage. The new concept of proactive network management aims at identifying the existing troubles in advance of any performance degradation, as well as providing support for future decision-making actions. The goal of this work is to adapt a complete proactive environment which will be used to explore the remote monitors and simulation tools to identify the symptoms of proactive management problems; and specially to recognize the problem using artificial intelligence techniques and to take proactive actions to solve it, configuring a proactive management application for the prevention of problems in computer networks.

This chapter describes the model speed and model in/out (I/O) efficiency of the high-resolution atmospheric general circulation model the Finite-volume Atmospheric Model of IAP/LASG (FAMIL), investigated at the National Supercomputer Center in Tianjin, China, on its Tianhe-1A supercomputer platform. A series of three-model-day simulations were conducted with the standard Aqua-Planet Experiment (APE) project, designed within FAMIL, to obtain time stamps for the calculation of model speed, simulation cost, and model I/O efficiency. The results of the simulation demonstrate that FAMIL has remarkable scalability below 3,456 and 6,144 cores; the lowest simulation costs were found to be 1,536 and 3,456 cores for 12.5 and 6.25 km resolutions, respectively. Furthermore, FAMIL has excellent I/O scalability and efficiencies of more than 80 and 99 % on 6 and 1,536 I/Os, respectively.