RGraph:基于RDMA的高效分布式图数据处理系统

图是描述实体间关系的重要数据结构,被广泛地应用于信息科学、物理学、生物学、环境生态学等重要的科学领域.现如今,随着图数据规模的不断增大,利用分布式系统来处理大图数据已经成为主流,出现了形如Pregel、GraphX、PowerGraph和Gemini等经典的分布式大图数据处理系统.然而,与当前先进的基于单机的图处理系统...     

L-PowerGraph: a lightweight distributed graph-parallel communication mechanism

In order to process complex and large-scale graph data, numerous distributed graph-parallel computing platforms have been proposed. PowerGraph is an excellent representative of them. It has exhibited better performance, such as faster graph-processing rate and higher scalability, than others. However, like in other distributed graph computing systems, unnecessary and excessive communications among computing nodes in PowerGraph not only aggravate the network I/O workload of the underlying computing hardware systems but may also cause a decrease in runtime performance. In this paper, we propose and implement a mechanism called L-PowerGraph, which reduces the communication overhead in PowerGraph. First, L-PowerGraph identifies and eliminates the avoidable communications in PowerGraph. Second, in order to further reduce the required communications L-PowerGraph proposes an edge direction-aware master appointment strategy, in which L-PowerGraph appoints the replica with both incoming and outgoing edges as master. Third, L-PowerGraph proposes an edge direction-aware graph partition strategy, which optimally isolates the outgoing edges from the incoming edges of a vertex during the graph partition process. We have conducted extensive experiments using real-world datasets, and our results verified the effectiveness of the proposed mechanism. For example, compared with PowerGraph under Random partition scenario L-PowerGraph can not only reduce up to 30.5% of the communication overhead but also cut up to 20.3% of the runtime for PageRank algorithm while processing Live-journal dataset. The performance improvement achieved by L-PowerGraph over our precursor work, LightGraph, which only reduces the synchronizing communication overhead, is also verified by our experimental results.

     

面向大规模二部图的分布式Tip分解算法

Tip分解作为图数据管理领域的热点研究问题,已被广泛应用于文档聚类和垃圾邮件组检测等实际场景中.随着图数据规模的爆炸式增长,单机内存已无法满足其存储需求,亟需研究分布式环境下Tip分解技术.现有分布式图计算系统的通信模式无法适用于二部图,为此,首先提出一种基于中继的通信模式,以实现分布式环境下处理二部图时消息的有效传递...     

Radius-aware approximate blank node matching using signatures

In the linked open data cloud, the biggest open data graph that currently exists, a remarkable percentage of data are unnamed resources, also called blank nodes. Several fundamental tasks, such as graph isomorphism checking and RDF data versioning, require computing a map between the sets of blank nodes of two graphs. This map aims at minimizing the delta size, i.e. the number of change operations that are required to make the graphs isomorphic. Computing the optimal map is NP-Hard in the general case, and various approximation algorithms have been proposed. In this work, we propose a novel radius-aware signature-based algorithm that is not restricted to the direct neighborhood of the compared blank nodes. Contrary to the older algorithms, the proposed algorithm manages to decrease the deviation from the optimal solution even for graphs that contain connected blank nodes in large and dense structures. The conducted experiments over real and synthetically generated datasets (including datasets from the Billion Triple Challenge 2012 and 2014) show the significantly smaller deltas. For isomorphism checking (simple RDF equivalence), with a wise configuration of radius, the proposed algorithm achieves optimality for \(100\,\%\) of the datasets, while in non-isomorphic datasets the deltas are on average 50–75 % smaller than those of the previous algorithms. Finally, the trade-off between radius, deviation from the optimum and time efficiency is analyzed.     

Distributed Tip Decomposition on Large Bipartite Graphs

Tip decomposition has a pivotal role in mining cohesive subgraphs in bipartite graphs. It is a popular research topic with wide applications in document clustering, spam group detection, and analysis of affiliation networks. With the explosive growth of the bipartite graph data scale in these scenarios, it is necessary to use distributed methods to realize its effective storage. For this reason, this paper studies the problem of the tip decomposition on a bipartite graph in the distributed environment for the first time. Firstly, a new relay-based communication mode is proposed to realize effective message transmission when the given bipartite graph is decomposed in a distributed environment. Secondly, the Distributed Butterfly Counting (DBC) algorithm and the Distributed Tip Decomposition (DTD) algorithm are designed. In particular, a controllable parallel vertex activation strategy is proposed to solve the problem of memory overflow when DBC decomposes large-scale bipartite graphs. Finally, the message pruning strategy based on vertex priority and message validity pruning strategy are introduced to further improve the efficiency of the algorithm by reducing redundant communication and computing overhead. The experiment is deployed on the high-performance distributed computing platform of the National Supercomputing Center. The effectiveness and efficiency of the proposed algorithms are verified by experiments on several real datasets.     

Scalable distributed Louvain algorithm for community detection in large graphs

Community detection (or clustering) in large-scale graphs is an important problem in graph mining. Communities reveal interesting organizational and functional characteristics of a network. Louvain algorithm is an efficient sequential algorithm for community detection. However, such sequential algorithms fail to scale for emerging large-scale data. Scalable parallel algorithms are necessary to process large graph datasets. In this work, we show a comparative analysis of our different parallel implementations of Louvain algorithm. We design parallel algorithms for Louvain method in shared memory and distributed memory settings. Developing distributed memory parallel algorithms is challenging because of inter-process communication and load balancing issues. We incorporate dynamic load balancing in our final algorithm DPLAL (Distributed Parallel Louvain Algorithm with Load-balancing). DPLAL overcomes the performance bottleneck of the previous algorithms and shows around 12-fold speedup scaling to a larger number of processors. We also compare the performance of our algorithm with some other prominent algorithms in the literature and get better or comparable performance . We identify the challenges in developing distributed memory algorithm and provide an optimized solution DPLAL showing performance analysis of the algorithm on large-scale real-world networks from different domains.

     

知识图谱划分算法研究综述

知识图谱是人工智能的重要基石,因其包含丰富的图结构和属性信息而受到广泛关注.知识图谱可以精确语义描述现实世界中的各种实体及其联系,其中顶点表示实体,边表示实体间的联系.知识图谱划分是大规模知识图谱分布式处理的首要工作,对知识图谱分布式存储、查询、推理和挖掘起基础支撑作用.随着知识图谱数据规模及分布式处理需求的不断增长,...     

An efficient parallel algorithm of N-hop neighborhoods on graphs in distributed environment

N-hop neighborhoods information is very useful in analytic tasks on large-scale graphs, like finding clique in a social network, recommending friends or advertising links according to one’s interests, predicting links among websites and etc. To get the N-hop neighborhoods information on a large graph, such as a web graph, a twitter social graph, the most straightforward method is to conduct a breadth first search (BFS) on a parallel distributed graph processing framework, such as Pregel and GraphLab. However, due to the massive volume of message transfer, the BFS method results in high communication cost and has low efficiency. In this work, we propose a key/value based method, namely KVB, which perfectly fits into the prevailing parallel graph processing framework and computes N-hop neighborhoods on a large scale graph efficiently. Unlike the BFS method, our method need not transfer large amount of neighborhoods information, thus, significantly reduces the overhead on both the communication and intermediate results in the distributed framework.We formalize the N-hop neighborhoods query processing as an optimization problem based on a set of quantitative cost metrics of parallel graph processing. Moreover, we propose a solution to efficiently load only the relevant neighborhoods for computation. Specially, we prove the optimal partial neighborhoods load problem is NP-hard and carefully design a heuristic strategy. We have implemented our algorithm on a distributed graph framework- Spark GraphX and validated our solution with extensive experiments over a number of real world and synthetic large graphs on a modest indoor cluster. Experiments show that our solution generally gains an order of magnitude speedup comparing to the state-of-art BFS implementation.     

An I/O-efficient and adaptive fault-tolerant framework for distributed graph computations

In recent year, many large-scale iterative graph computation systems such as Pregel have been developed. To ensure that these systems are fault-tolerant, checkpointing, which archives graph states onto distributed file systems periodically, has been proposed. However, fault-tolerance remains to be challenging because the whole data set is archived with a static interval, rendering underlying graph computations to entail I/O-costs in terms of disk and network communication. Motivated by this, we first propose to dynamically adjust checkpoint intervals based on a carefully designed cost-analysis model, by taking the underlying computing workload into account. Furthermore, for algorithms that can be restarted from any point during computations, we prioritize graph states and then checkpointing can be performed with selected data, instead of the entire dataset, to reduce archiving overhead while simultaneously guaranteeing the failure recovery efficiency. Finally, we conduct extensive performance studies to confirm the effectiveness of our approaches over existing up-to-date solutions using a broad spectrum of real-world graphs.     

SOOP:Efficient Distributed Graph Computation Supporting Second-Order Random Walks

The second-order random walk has recently been shown to effectively improve the accuracy in graph analysis tasks.Existing work mainly focuses on centralized second-order random walk (SOW) algorithms.SOW algorithms rely on edge-to-edge transition probabilities to generate next random steps.However,it is prohibitively costly to store all the probabilities for large-scale graphs,and restricting the number of probabilities to consider can negatively impact the accuracy of graph analysis tasks.In this paper,we propose and study an alternative approach,SOOP (second-order random walks with on-demand probability computation),that avoids the space overhead by computing the edge-to-edge transition probabilities on demand during the random walk.However,the same probabilities may be computed multiple times when the same edge appears multiple times in SOW,incurring extra cost for redundant computation and communication.We propose two optimization techniques that reduce the complexity of computing edge-to-edge transition probabilities to generate next random steps,and reduce the cost of communicating out-neighbors for the probability computation,respectively.Our experiments on real-world and synthetic graphs show that SOOP achieves orders of magnitude better performance than baseline precompute solutions,and it can efficiently computes SOW algorithms on billion-scale graphs.     

Mining collective knowledge: inferring functional labels from online review for business

Uncertain graph has been widely used to represent graph data with inherent uncertainty in structures. Reliability search is a fundamental problem in uncertain graph analytics. This paper investigates on a new problem with broad real-world applications, the top-k reliability search problem on uncertain graphs, that is, finding the k vertices v with the highest reliabilities of connections from a source vertex s to v. Note that the existing algorithm for the threshold-based reliability search problem is inefficient for the top-k reliability search problem. We propose a new algorithm to efficiently solve the top-k reliability search problem. The algorithm adopts two important techniques, namely the BFS sharing technique and the offline sampling technique. The BFS sharing technique exploits overlaps among different sampled possible worlds of the input uncertain graph and performs a single BFS on all possible worlds simultaneously. The offline sampling technique samples possible worlds offline and stores them using a compact structure. The algorithm also takes advantages of bit vectors and bitwise operations to improve efficiency. In addition, we generalize the top-k reliability search problem from single-source case to the multi-source case and show that the multi-source case of the problem can be equivalently converted to the single-source case of the problem. Moreover, we define two types of the reverse top-k reliability search problems with different semantics on uncertain graphs. We propose appropriate solutions for both of them. Extensive experiments carried out on both real and synthetic datasets verify that the optimized algorithm outperforms the baselines by 1–2 orders of magnitude in execution time while achieving comparable accuracy. Meanwhile, the optimized algorithm exhibits linear scalability with respect to the size of the input uncertain graph.     

NScale: neighborhood-centric large-scale graph analytics in the cloud

There is an increasing interest in executing complex analyses over large graphs, many of which require processing a large number of multi-hop neighborhoods or subgraphs. Examples include ego network analysis, motif counting, finding social circles, personalized recommendations, link prediction, anomaly detection, analyzing influence cascades, and others. These tasks are not well served by existing vertex-centric graph processing frameworks, where user programs are only able to directly access the state of a single vertex at a time, resulting in high communication, scheduling, and memory overheads in executing such tasks. Further, most existing graph processing frameworks ignore the challenges in extracting the relevant portions of the graph that an analysis task is interested in, and loading those onto distributed memory. This paper introduces NScale, a novel end-to-end graph processing framework that enables the distributed execution of complex subgraph-centric analytics over large-scale graphs in the cloud. NScale enables users to write programs at the level of subgraphs rather than at the level of vertices. Unlike most previous graph processing frameworks, which apply the user program to the entire graph, NScale allows users to declaratively specify subgraphs of interest. Our framework includes a novel graph extraction and packing (GEP) module that utilizes a cost-based optimizer to partition and pack the subgraphs of interest into memory on as few machines as possible. The distributed execution engine then takes over and runs the user program in parallel on those subgraphs, restricting the scope of the execution appropriately, and utilizes novel techniques to minimize memory consumption by exploiting overlaps among the subgraphs. We present a comprehensive empirical evaluation comparing against three state-of-the-art systems, namely Giraph, GraphLab, and GraphX, on several real-world datasets and a variety of analysis tasks. Our experimental results show orders-of-magnitude improvements in performance and drastic reductions in the cost of analytics compared to vertex-centric approaches.     

A parallel fuzzy clustering algorithm for large graphs using Pregel

Large graphs are scale free and ubiquitous having irregular relationships. Clustering is used to find existent similar patterns in graphs and thus help in getting useful insights. In real-world, nodes may belong to more than one cluster thus, it is essential to analyze fuzzy cluster membership of nodes. Traditional centralized fuzzy clustering algorithms incur high communication cost and produce poor quality of clusters when used for large graphs. Thus, scalable solutions are obligatory to handle huge amount of data in less computational time with minimum disk access. In this paper, we proposed a parallel fuzzy clustering algorithm named ‘PGFC’ for handling scalable graph data. It will be advantageous from the viewpoint of expert systems to develop a clustering algorithm that can assure scalability along with better quality of clusters for handling large graphs.The algorithm is parallelized using bulk synchronous parallel (BSP) based Pregel model. The cluster centers are initialized using degree centrality measure, resulting in lesser number of iterations. The performance of PGFC is compared with other state of art clustering algorithms using synthetic graphs and real world networks. The experimental results reveal that the proposed PGFC scales up linearly to handle large graphs and produces better quality of clusters when compared to other graph clustering counterparts.     

基于顶点组重分配的动态增量图划分算法

图划分是分布式图计算中的一项基础工作, 其作用是将大规模图进行划分并分配到集群中的不同机器上. 图划分的质量对分布式图计算的性能有很大的影响, 其目标是降低负载平衡和最小化边割. 如今, 现实中的图数据通常呈动态增长态势, 这就需要一种能够处理动态增量图的划分方法, 在图数据动态增长的过程中确保划分的质量不受影响. 目前虽然有一些动态图划分算法被提出, 但它们不能同时专注于实时处理动态变化和获得高质量的划分结果. 提出基于顶点组重分配的动态增量图划分算法(ED-IDGP)来解决大规模动态增量图的划分问题. 在ED-IDGP算法中, 设计实时处理4种不同单元更新类型的动态处理器, 并在每次处理完单元更新后通过在分区发生动态变化的附近执行局部优化器进一步提高图划分的质量. 在ED-IDGP的局部优化器中, 利用基于改进标签传播算法的顶点组搜索策略搜索顶点组, 并利用提出的顶点组移动增益公式衡量最有益的顶点组, 将该顶点组移动到目标分区中做优化. 在真实数据集上从不同的角度和度量指标评估了ED-IDGP算法的性能和效率.     

A communication-reduced and computation-balanced framework for fast graph computation

The bulk synchronous parallel (BSP) model is very user friendly for coding and debugging parallel graph algorithms. However, existing BSP-based distributed graph-processing frameworks, such as Pregel, GPS and Giraph, routinely suffer from high communication costs. These high communication costs mainly stem from the fine-grained message-passing communication model. In order to address this problem, we propose a new computation model with low communication costs, called LCC-BSP. We use this model to design and implement a high-performance distributed graph-processing framework called LCC-Graph. This framework eliminates high communication costs in existing distributed graph-processing frameworks. Moreover, LCC-Graph also balances the computation workloads among all compute nodes by optimizing graph partitioning, significantly reducing the computation time for each superstep. Evaluation of LCC-Graph on a 32-node cluster, driven by real-world graph datasets, shows that it significantly outperforms existing distributed graph-processing frameworks in terms of runtime, particularly when the system is supported by a high-bandwidth network. For example, LCC-Graph achieves an order of magnitude performance improvement over GPS and GraphLab.     

Efficient parallel edge-centric approach for relaxed graph pattern matching

Prior algorithms on graph simulation for distributed graphs are not scalable enough as they exhibit heavy message passing. Moreover, they are dependent on the graph partitioning quality that can be a bottleneck due to the natural skew present in real-world data. As a result, their degree of parallelism becomes limited. In this paper, we propose an efficient parallel edge-centric approach for distributed graph pattern matching. We design a novel distributed data structure called ST that allows a fine-grain parallelism, and hence guarantees linear scalability. Based on ST, we develop a parallel graph simulation algorithm called PGSim. Furthermore, we propose PDSim, an edge-centric algorithm that efficiently evaluates dual simulation in parallel. PDSim combines ST and PGSim in a Split-and-Combine approach to accelerate the computation stages. We prove the effectiveness and efficiency of these propositions through theoretical guarantees and extensive experiments on massive graphs. The achieved results confirm that our approach outperforms existing algorithms by more than an order of magnitude.

     

Efficient computation of distance labeling for decremental updates in large dynamic graphs

Since today’s real-world graphs, such as social network graphs, are evolving all the time, it is of great importance to perform graph computations and analysis in these dynamic graphs. Due to the fact that many applications such as social network link analysis with the existence of inactive users need to handle failed links or nodes, decremental computation and maintenance for graphs is considered a challenging problem. Shortest path computation is one of the most fundamental operations for managing and analyzing large graphs. A number of indexing methods have been proposed to answer distance queries in static graphs. Unfortunately, there is little work on answering such queries for dynamic graphs. In this paper, we focus on the problem of computing the shortest path distance in dynamic graphs, particularly on decremental updates (i.e., edge deletions). We propose maintenance algorithms based on distance labeling, which can handle decremental updates efficiently. By exploiting properties of distance labeling in original graphs, we are able to efficiently maintain distance labeling for new graphs. We experimentally evaluate our algorithms using eleven real-world large graphs and confirm the effectiveness and efficiency of our approach. More specifically, our method can speed up index re-computation by up to an order of magnitude compared with the state-of-the-art method, Pruned Landmark Labeling (PLL).     

GPPR: Geo-distributed Personalized PageRank Algorithm

Personalized PageRank, as a basic algorithm in large graph analysis, has a wide range of applications in search engines, social recommendation, community detection, and other fields and it has been a hot problem of interest to researchers. The existing distributed personalized PageRank algorithms assume that all data are located in the same geographic location and the network environment is the same among the computing nodes where the data are located. However, in the real world, these data may be distributed in multiple data centers across continents, and these geo-distributed data centers are connected to each other through WANs, which are characterized by heterogeneous network bandwidth, huge hardware differences, and high communication costs. Moreover, the distributed personalized PageRank algorithm requires multiple iterations and random walk on the global graph. Therefore, the existing distributed personalized PageRank algorithms are not applicable to the geo-distributed environment. To address this problem, the GPPR (Geo-distributed Personalized PageRank) algorithm is proposed in this paper. The algorithm first preprocesses the big graph data in the geo-distributed environment and maps the graph data by using a heuristic algorithm to reduce the impact of network bandwidth heterogeneity on the iteration speed of the algorithm. Secondly, GPPR improves the random walk approach and proposes a probability-based push algorithm to further lower the number of iterations required by the algorithm by reducing the bandwidth load of data transmission between working nodes. We implement the GPPR algorithm based on the Spark framework and build a real geo-distributed environment in AliCloud to conduct experiments comparing the GPPR algorithm with several existing representative distributed personalized PageRank algorithms on eight open-source big graph datasets. The results show that the communication data volume of GPPR is reduced by 30% on average in the geo-distributed environment compared with that of other algorithms. In terms of algorithm running efficiency, GPPR improves by an average 2.5 factor compared with other algorithms.     

Partitioning dynamic graph asynchronously with distributed FENNEL

Graph partitioning is important in distributed graph processing. Classical method such as METIS works well on relatively small graphs, but hard to scale for huge, dynamic graphs. Streaming graph partitioning algorithms overcome this issue by processing those graphs as streams. Among these algorithms, FENNEL achieves better edge cut ratio, even close to METIS, but consumes less memory and is significantly faster. On the other hand, graph partitioning may also benefit from distributed graph processing. However, to deploy FENNEL on a cluster, it is important to avoid quality loss and keep efficiency high. The direct implementation of this idea yields a synchronous model and a star-shaped network, which limits both scalability and efficiency. Targeting these two problems, we propose an asynchronous model, combined with a dedicated tree-shaped map-reduce network which is prevail in systems such as Apache Hadoop and Spark, to form AsyncFENNEL (asynchronous FENNEL). We theoretically prove that, the impact on partition quality brought by asynchronous model can be kept as minimal. We test AsyncFENNEL with various synthetic and real-world graphs, the comparison between synchronous and asynchronous models shows that, for streamed natural graphs, AsyncFENNEL can improve performance significantly (above 300% with 8 workers/partitions) with negligible loss on edge cut ratio. However, more worker nodes will introduce a heavier network traffic and reduce efficiency. The proposed tree-shaped map-reduce network can mitigate that impact and increase the performance in that case.     

A study of graph partitioning schemes for parallel graph community detection

This paper presents a study of graph partitioning schemes for parallel graph community detection on distributed memory machines. We investigate the relationship between graph structure and parallel clustering effectiveness, and develop a heuristic partitioning algorithm suitable for modularity-based algorithms. We demonstrate the accuracy and scalability of our approach using several real-world large graph datasets compared with state-of-the-art parallel algorithms on the Cray XK7 supercomputer at Oak Ridge National Laboratory. Given the ubiquitous graph model, we expect this high-performance solution will help lead to new insights in numerous fields.