What is Twitter

Traditionally distributed graph processing systems have largely focused on scalability through the optimizations of inter-node communication and load balance. However, they often deliver unsatisfactory overall processing efficiency compared with shared-memory graph computing frameworks. We analyze the behavior of several graph-parallel systems and find that the added overhead for achieving scalability becomes a major limiting factor for efficiency, especially with modern multi-core processors and high-speed interconnection networks.Based on our observations, we present Gemini, a distributed graph processing system that applies multiple optimizations targeting computation performance to build scalability on top of efficiency. Gemini adopts (1) a sparse-dense signal-slot abstraction to extend the hybrid push-pull computation model from shared-memory to distributed scenarios, (2) a chunk-based partitioning scheme enabling low-overhead scaling out designs and locality-preserving vertex accesses, (3) a dual representation scheme to compress accesses to vertex indices, (4) NUMA-aware sub-partitioning for efficient intra-node memory accesses, plus (5) locality-aware chunking and fine-grained work-stealing for improving both internode and intra-node load balance, respectively. Our evaluation on an 8-node high-performance cluster (using five widely used graph applications and five real-world graphs) shows that Gemini significantly outperforms all well-known existing distributed graph processing systems, delivering up to 39.8× (from 8.91×) improvement over the fastest among them.

/pdf/gemini-a-computation-centric-distributed-graph-processing-57yv5aorjj.pdf

Gemini: a computation-centric distributed graph processing system

In this paper, we present GridGraph, a system for processing large-scale graphs on a single machine. Grid-Graph breaks graphs into 1D-partitioned vertex chunks and 2D-partitioned edge blocks using a first fine-grained level partitioning in preprocessing. A second coarse-grained level partitioning is applied in runtime. Through a novel dual sliding windows method, GridGraph can stream the edges and apply on-the-fly vertex updates, thus reduce the I/O amount required for computation. The partitioning of edges also enable selective scheduling so that some of the blocks can be skipped to reduce unnecessary I/O. This is very effective when the active vertex set shrinks with convergence.

Our evaluation results show that GridGraph scales seamlessly with memory capacity and disk bandwidth, and outperforms state-of-the-art out-of-core systems, including GraphChi and X-Stream. Furthermore, we show that the performance of GridGraph is even competitive with distributed systems, and it also provides significant cost efficiency in cloud environment.

/pdf/gridgraph-large-scale-graph-processing-on-a-single-machine-2t1iqmbo27.pdf

GridGraph: large-scale graph processing on a single machine using 2-level hierarchical partitioning

Natural graphs with skewed distribution raise unique challenges to graph computation and partitioning. Existing graph-parallel systems usually use a "one size fits all" design that uniformly processes all vertices, which either suffer from notable load imbalance and high contention for high-degree vertices (e.g., Pregel and GraphLab), or incur high communication cost and memory consumption even for low-degree vertices (e.g., PowerGraph and GraphX). In this paper, we argue that skewed distribution in natural graphs also calls for differentiated processing on high-degree and low-degree vertices. We then introduce PowerLyra, a new graph computation engine that embraces the best of both worlds of existing graph-parallel systems, by dynamically applying different computation and partitioning strategies for different vertices. PowerLyra further provides an efficient hybrid graph partitioning algorithm (hybrid-cut) that combines edge-cut and vertex-cut with heuristics. Based on PowerLyra, we design locality-conscious data layout optimization to improve cache locality of graph accesses during communication. PowerLyra is implemented as a separate computation engine of PowerGraph, and can seamlessly support various graph algorithms. A detailed evaluation on two clusters using graph-analytics and MLDM (machine learning and data mining) applications show that PowerLyra outperforms PowerGraph by up to 5.53X (from 1.24X) and 3.26X (from 1.49X) for real-world and synthetic graphs accordingly, and is much faster than other systems like GraphX and Giraph, yet with much less memory consumption. A porting of hybrid-cut to GraphX further confirms the efficiency and generality of PowerLyra.

/pdf/powerlyra-differentiated-graph-computation-and-partitioning-2urz40n275.pdf

PowerLyra: differentiated graph computation and partitioning on skewed graphs

The vertex-centric programming model is an established computational paradigm recently incorporated into distributed processing frameworks to address challenges in large-scale graph processing. Billion-node graphs that exceed the memory capacity of commodity machines are not well supported by popular Big Data tools like MapReduce, which are notoriously poor performing for iterative graph algorithms such as PageRank. In response, a new type of framework challenges one to “think like a vertex” (TLAV) and implements user-defined programs from the perspective of a vertex rather than a graph. Such an approach improves locality, demonstrates linear scalability, and provides a natural way to express and compute many iterative graph algorithms. These frameworks are simple to program and widely applicable but, like an operating system, are composed of several intricate, interdependent components, of which a thorough understanding is necessary in order to elicit top performance at scale. To this end, the first comprehensive survey of TLAV frameworks is presented. In this survey, the vertex-centric approach to graph processing is overviewed, TLAV frameworks are deconstructed into four main components and respectively analyzed, and TLAV implementations are reviewed and categorized.

https://dl.acm.org/doi/pdf/10.1145/2818185

Thinking Like a Vertex: A Survey of Vertex-Centric Frameworks for Large-Scale Distributed Graph Processing

Graph-structured analytics has been widely adopted in a number of big data applications such as social computation, web-search and recommendation systems. Though much prior research focuses on scaling graph-analytics on distributed environments, the strong desire on performance per core, dollar and joule has generated considerable interests of processing large-scale graphs on a single server-class machine, which may have several terabytes of RAM and 80 or more cores. However, prior graph-analytics systems are largely neutral to NUMA characteristics and thus have suboptimal performance. This paper presents a detailed study of NUMA characteristics and their impact on the efficiency of graph-analytics. Our study uncovers two insights: 1) either random or interleaved allocation of graph data will significantly hamper data locality and parallelism; 2) sequential inter-node (i.e., remote) memory accesses have much higher bandwidth than both intra- and inter-node random ones. Based on them, this paper describes Polymer, a NUMA-aware graph-analytics system on multicore with two key design decisions. First, Polymer differentially allocates and places topology data, application-defined data and mutable runtime states of a graph system according to their access patterns to minimize remote accesses. Second, for some remaining random accesses, Polymer carefully converts random remote accesses into sequential remote accesses, by using lightweight replication of vertices across NUMA nodes. To improve load balance and vertex convergence, Polymer is further built with a hierarchical barrier to boost parallelism and locality, an edge-oriented balanced partitioning for skewed graphs, and adaptive data structures according to the proportion of active vertices. A detailed evaluation on an 80-core machine shows that Polymer often outperforms the state-of-the-art single-machine graph-analytics systems, including Ligra, X-Stream and Galois, for a set of popular real-world and synthetic graphs.

/pdf/numa-aware-graph-structured-analytics-4qag05fs2t.pdf

NUMA-aware graph-structured analytics

The increasing algorithm complexity and dataset sizes necessitate the use of networked machines for many graph-parallel algorithms, which also makes fault tolerance a must due to the increasing scale of machines. Unfortunately, existing large-scale graph-parallel systems usually adopt a distributed checkpoint mechanism for fault tolerance, which incurs not only notable performance overhead but also lengthy recovery time. This paper observes that the vertex replicas created for distributed graph computation can be naturally extended for fast in-memory recovery of graph states. This paper proposes Imitator, a new fault tolerance mechanism, that supports cheaply maintenance of vertex states by replicating vertex states to their replicas during normal message exchanges, and provides fast in-memory reconstruction of failed vertices from replicas in other machines. Imitator has been implemented by extending Hama, a popular open-source clone of Pregel. Evaluation shows that Imitator incurs negligible performance overhead (less than 5% for all cases) and can recover from failures of more than one million of vertices with less than 3.4 seconds.

Replication-Based Fault-Tolerance for Large-Scale Graph Processing

The increasing algorithmic complexity and dataset sizes necessitate the use of networked machines for many graph-parallel algorithms, which also makes fault tolerance a must due to the increasing scale of machines Unfortunately, existing large-scale graph-parallel systems usually adopt a distributed checkpoint mechanism for fault tolerance, which incurs not only notable performance overhead but also lengthy recovery time This paper observes that the vertex replicas created for distributed graph computation can be naturally extended for fast in-memory recovery of graph states This paper describes Imitator, a new fault tolerance mechanism, which supports cheap maintenance of vertex states by replicating them to their replicas during normal message exchanges, and provides fast in-memory reconstruction of failed vertices from replicas in other machines Imitator has been implemented on Cyclops with edge-cut and PowerLyra with vertex-cut Evaluation on a 50-node EC-2 like cluster shows that Imitator incurs an average of 137 and 232 percent performance overhead (ranging from −06 to 37 percent) for Cyclops and PowerLyra respectively, and can recover from failures of more than one million of vertices with less than 34 seconds

/pdf/replication-based-fault-tolerance-for-large-scale-graph-3obb7odxdl.pdf

Kaiyuan Zhang

Papers

NUMA-aware graph-structured analytics

Replication-Based Fault-Tolerance for Large-Scale Graph Processing

Replication-Based Fault-Tolerance for Large-Scale Graph Processing