DepGraph: A Dependency-Driven Accelerator for Efficient Iterative Graph Processing

Zhang, Yu; Liao, Xiaofei; Jin, Hai; He, Ligang; He, Bingsheng; Liu, Haikun; Gu, Lin

doi:10.1109/hpca51647.2021.00039

Cited by 22 publications

(14 citation statements)

References 65 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Note that, for different graph datasets, the memory overhead required by TDGraph for maintaining the extra data structures (e.g., Coalesced_States, H_Table, and Topology_List) accounts for 0.3%∼2.4% of the total memory consumption. Meanwhile, we also integrate Ligra-o with four cutting-edge graph processing accelerators, i.e., HATS [36], Minnow [67], PHI [37], and DepGraph [73], respectively. After that, TDGraph-H is compared with these accelerators.…”

Section: Evaluation 41 Experimental Methodologymentioning

confidence: 99%

“…Compared with the state-of-the-art software streaming graph processing system, TDGraph achieves the speedup of 7.1∼21.4 times with only 0.73% area cost. Compared with four cutting-edge hardware accelerators, i.e., HATS [36], Minnow [67], PHI [37], and DepGraph [73], TDGraph improves the performance by 4.6∼12.7, 3.2∼8.6, 3.8∼9.7, and 2.3∼6.1 times, respectively.…”

Section: Figure 1: Illustration Of Streaming Graph Analyticsmentioning

confidence: 99%

“…Meanwhile, some hardware prefetchers [5,6,10,66,74] are also designed to hide graph data access latency. For achieving faster state propagation and better data locality, several hardware prefetchers, e.g., HATS [36], Minnow [67], and DepGraph [73], have been further proposed. Nevertheless, when using these hardware solutions to support streaming graph processing, they still suffer from significant redundant computation and data access overhead due to irregular state propagations of the vertices affected by the graph updates (see Section 4.3).…”

Section: Hardware Support For Graph Processingmentioning

confidence: 99%

“…5 shows that each core of the many-core processor is coupled with a TDGraph engine, which resides between this core and its private L2 cache. Like prior works [36,67,73], TDGraph operates on virtual addresses and accesses the graph data through the L2 TLB and L2 cache of its paired core. The data will be retrieved from the LLC, or from the main memory, when the data are not in the L2 cache.…”

Section: Overview Of Tdgraphmentioning

confidence: 99%

“…Although some hardware solutions [9,19,36,37,64,67,73] have been designed to augment the many-core processors for highperformance graph processing, they are designed to accelerate the processing of static graphs. Meanwhile, to support streaming graph processing, some ASIC-based accelerators, e.g., JetStream [44] and DREDGE [35], have been recently proposed.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

TDGraph

Zhao

Yang

Liao

et al. 2022

Proceedings of the 49th Annual International Symposium on Computer Architecture

Self Cite

View full text Add to dashboard Cite

Many solutions have been recently proposed to support the processing of streaming graphs. However, for the processing of each graph snapshot of a streaming graph, the new states of the vertices affected by the graph updates are propagated irregularly along the graph topology. Despite the years' research efforts, existing approaches still suffer from the serious problems of redundant computation overhead and irregular memory access, which severely underutilizes a many-core processor. To address these issues, this paper proposes a topology-driven programmable accelerator TDGraph, which is the first accelerator to augment the many-core processors to achieve high performance processing of streaming graphs. Specifically, we propose an efficient topology-driven incremental execution approach into the accelerator design for more regular state propagation and better data locality. TDGraph takes the vertices affected by graph updates as the roots to prefetch other vertices along the graph topology and synchronizes the incremental computations of them on the fly. In this way, most state propagations originated from multiple vertices affected by different graph updates can be conducted together along the graph topology, which help reduce the redundant computations and data access cost. Besides, through the efficient coalescing of the accesses to vertex states, TDGraph further improves the utilization of the cache and memory bandwidth. We have evaluated TDGraph on a simulated 64-core processor. The results show that, the state-of-the-art software system achieves the speedup of 7.1∼21.4 times after integrating with TDGraph, while incurring only 0.73% area cost. Compared with four cutting-edge accelerators, i.e., HATS, Minnow, PHI, and DepGraph, TDGraph gains the speedups of 4.6∼12.7, 3.2∼8.6, 3.8∼9.7, and 2.3∼6.1 times, respectively.

show abstract

Section: Evaluation 41 Experimental Methodologymentioning

confidence: 99%

Section: Figure 1: Illustration Of Streaming Graph Analyticsmentioning

confidence: 99%