Gluon-Async: A Bulk-Asynchronous System for Distributed and Heterogeneous Graph Analytics

Dathathri, Roshan; Gill, Gurbinder; Hoang, Loc; Jatala, Vishwesh; Pingali, Keshav; Nandivada, V. Krishna; Dang, Hoang-Vu; Snir, Marc

doi:10.1109/pact.2019.00010

Cited by 19 publications

(10 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The generalized minimum spanning tree computation approach [15], [16] may appear attractive because of its simplicity and greater work efficiency, however, we observe that it is possible to develop an algorithm with higher parallel efficiency for Voronoi cell computation compared to that of MST computation. Bader et al [18] and authors of the Galois [21] project demonstrated MST computation suffers from rapid decrease in the available parallelism [22].…”

Section: Parallel Steiner Tree Algorithmmentioning

confidence: 99%

“…(ii) offers load balancing for scale-free graphs through vertexcut partitioning by distributing edges of high-degree vertices across multiple partitions -crucial to scale to large graphs with skewed degree distribution; and (iii) an MPI-based implementation is likely more efficient than a Hadoop/Spark based solution [28]. Note that our solution can easily be implemented by extending other general purpose graph processing frameworks that expose a vertex-centric API, e.g., Gluon-Galois [21], GraphLab, Giraph, and GraphX [23], independent of BSP or asynchronous processing approaches.…”

Section: Distributed Implementationmentioning

confidence: 99%

“…For each graph, we create symmetric edges (2|E| edges) with non-zero, positive edge weights in the range as listed in Table III. The Web Data Commons 2012 (WDC), ClueWeb 2012 (CLW), and UK Web 2007-05 (UKW) are web graphs whose vertices are webpages and edges are hyperlinks [21]. Friendster (FSR) [21] and LiveJournal (LVJ) [30] are usercentric social media graphs.…”

Section: Distributed Implementationmentioning

confidence: 99%

See 2 more Smart Citations

Towards Distributed 2-Approximation Steiner Minimal Trees in Billion-edge Graphs

Reza¹,

Sanders²,

Pearce³

2022

Preprint

View full text Add to dashboard Cite

Given an edge-weighted graph and a set of known seed vertices of interest, a network scientist often desires to understand the graph relationships to explain connections between the seed vertices. If the size of the seed set is 2, shortest path calculations are an attractive computational kernel to explore the connections between the two vertices. When the seed set is 3 or larger (say up to 1,000s) Steiner minimal tree -min-weight acyclic connected subgraph (of the input graph) that contains all the seed vertices -is an attractive generalization of shortest weighted paths. In general, computing a Steiner minimal tree is NP-hard, but decades ago several polynomial-time algorithms were designed and proven to yield Steiner trees whose total weight is bounded within 2 times the minimal Steiner tree.Despite its rich theoretical literature, works related to parallel Steiner minimal tree computation and their scalable implementations are rather scarce. In this paper, we present a parallel 2approximation Steiner minimal tree algorithm (with theoretical guarantees) and its MPI-based distributed implementation. In place of distance computation between all pairs of seed vertices, an expensive phase in many approximation algorithms, the solution we employ, exploits Voronoi cell computation. Also, this approach has higher parallel efficiency than others that involve minimum spanning tree computation on the entire graph. Furthermore, our distributed design exploits asynchronous processing and a message prioritization scheme to accelerate convergence of distance computation, employs techniques to avoid inefficient distributed spanning tree computation on the entire graph, and harnesses a combination of vertex and edge centric processing to offer fast time-to-solution. We demonstrate scalability and performance of our solution using real-world graphs with up to 128 billion edges and 512 compute nodes (8K processes), show the ability to find Steiner trees with up to 10K seed vertices in under one minute, and present in-depth analyses that highlight the benefits of our design choices. Using four real-world graphs and three seed sets for each, we compare our solution with the state-of-the-art exact Steiner minimal tree solver, SCIP-Jack, and two sequential algorithms with the same approximation bound as our algorithm. Our distributed solution comfortably outperforms these related works on graphs with 10s million edges and offers decent strong scaling -up to 90% efficient. We empirically show that, on average, the total distance (sum of edge weights) of the Steiner tree identified by our solution is 1.0527 times greater than the Steiner minimal tree (i.e., the optimal solution) -well within the theoretical bound of less than equal to 2.

show abstract

Section: Parallel Steiner Tree Algorithmmentioning

confidence: 99%

Section: Distributed Implementationmentioning

confidence: 99%

Section: Distributed Implementationmentioning

confidence: 99%

See 1 more Smart Citation

Towards Distributed 2-Approximation Steiner Minimal Trees in Billion-edge Graphs

Reza¹,

Sanders²,

Pearce³

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Yet, both Groute and Gunrock show limited scalability going from one GPU to multiple GPUs. A more recent study by Jatala et al [31] shows that the performance of several distributed frameworks for BFS, including the asynchronous Gluon [17], across multiple configurations. The key execution bottleneck for these algorithms is communication, which accounts for roughly 70% of the total execution time.…”

Section: Related Workmentioning

confidence: 99%

ButterFly BFS -- An Efficient Communication Pattern for Multi Node Traversals

Green¹

2021

Preprint

View full text Add to dashboard Cite

Breadth-First Search (BFS) is a building block used in a wide array of graph analytics and is used in various network analysis domains: social, road, transportation, communication, and much more. Over the last two decades, network sizes have continued to grow. The popularity of BFS has brought with it a need for significantly faster traversals. Thus, BFS algorithms have been designed to exploit shared-memory and shared-nothing systems-this includes algorithms for accelerators such as the GPU. GPUs offer extremely fast traversals at the cost of processing smaller graphs due to their limited memory size. In contrast, CPU shared-memory systems can scale to graphs with several billion edges but do not have enough compute resources needed for fast traversals. This paper introduces ButterFly BFS, a multi-GPU traversal algorithm that allows analyzing significantly larger networks at high rates. ButterFly BFS scales to the similar-sized graphs processed by shared-memory systems while improving performance by more than 10X compared to CPUs. We evaluate our new algorithm on an NVIDIA DGX-2 server with 16 V100 GPUS and show that our algorithm scales with an increase in the number of GPUS. We show that we can achieve a roughly 70% performance linear speedup, which is non-trivial for BFS. For a scale 29 Kronecker graph and edge factor of 8, our new algorithm traverses the graph at a rate of over 300 GTEP/s. That is a high traversal rate for a single server.

show abstract

“…Similar asynchronous execution is explored on single-node shared memory systems using thread-local work queues [11] and bulk-asynchronous parallel execution on heterogeneous systems [12]. These works focus on asynchronous execution, without considering the benefits of hybrid synchronousasynchronous behavior.…”

Section: Background and Motivation A Synchrony And Asynchrony In Grap...mentioning

confidence: 99%

Delayed Asynchronous Iterative Graph Algorithms

Blanco¹,

McMillan²,

Low³

2021

Preprint

View full text Add to dashboard Cite

Iterative graph algorithms often compute intermediate values and update them as computation progresses. Updated output values are used as inputs for computations in current or subsequent iterations; hence the number of iterations required for values to converge can potentially reduce if the newest values are asynchronously made available to other updates computed in the same iteration. In a multi-threaded shared memory system, the immediate propagation of updated values can cause memory contention that may offset the benefit of propagating updates sooner. In some cases, the benefit of a smaller number of iterations may be diminished by each iteration taking longer.Our key idea is to combine the low memory contention that synchronous approaches have with the faster information sharing of asynchronous approaches. Our hybrid approach buffers updates from threads locally before committing them to the global store to control how often threads may cause conflicts for others while still sharing data within one iteration and hence speeding convergence. On a 112-thread CPU system, our hybrid approach attains up to 4.5% -19.4% speedup over an asynchronous approach for Pagerank and up to 1.9% -17% speedup over asynchronous Bellman Ford SSSP. Further, our hybrid approach attains 2.56x better performance than the synchronous approach. Finally, we provide insights as to why delaying updates is not helpful on certain graphs where connectivity is clustered on the main diagonal of the adjacency matrix.

show abstract

Gluon-Async: A Bulk-Asynchronous System for Distributed and Heterogeneous Graph Analytics

Cited by 19 publications

References 29 publications

Towards Distributed 2-Approximation Steiner Minimal Trees in Billion-edge Graphs

Towards Distributed 2-Approximation Steiner Minimal Trees in Billion-edge Graphs

ButterFly BFS -- An Efficient Communication Pattern for Multi Node Traversals

Delayed Asynchronous Iterative Graph Algorithms

Contact Info

Product

Resources

About