AccuGraph normalized to the ForeGraph performance[9]. YT: Youtube; Wk: Wiki; LJ: LiveJournal; AVG: average; PR: PageRank; WCC: Weakly Connected Components.

Loading [MathJax]/jax/output/SVG/jax.js

Download： Full Size Img PowerPoint

AccuGraph normalized to the ForeGraph performance^[9]. YT: Youtube; Wk: Wiki; LJ: LiveJournal; AVG: average; PR: PageRank; WCC: Weakly Connected Components.

2024, 39(2): 245-266.

Towards High-Performance Graph Processing: From a Hardware/Software Co-Design Perspective .

Figures of the Article

Graph processing ecosystem. BFS: Breath First Search; SSSP: Single Source Shortest Path.
(a) Architecture of AccuGraph and (b) detailed design of the parallel accumulator^[9]. P_i denotes the i-th pipeline stage. Acc. means accumulator.
AccuGraph normalized to the ForeGraph performance^[9]. YT: Youtube; Wk: Wiki; LJ: LiveJournal; AVG: average; PR: PageRank; WCC: Weakly Connected Components.
Overall architecture of ScalaBFS2. PG: processing group; LT: local store.
ScalaGraph architecture^[13]. Pref: prefetcher, VPref: vertex prefetcher, EPref: edge prefetcher, EDU: edge dispathing unit, DU: dispathing unit, SPD: scratchpad, RU: routing unit, GU: graph unit, and VDU: vertex dispatching unit.
Throughput^[13] of ScalaGraph vs GraphDyns and Gunrock. CC: connected components; PK: Pokec; OR: Orkut; RM: RMAT^[24]; TW: Twitter.
Example illustrating how to extract a hot graph^[18]. (a) Depth-first search. (b) Partitions generated.
Redundant memory accesses ratio^[19]. UK: UKDomin; TT: Twitter MPI; FT: Friendster.
Example of how to partition a directed graph into directed paths and obtain a dependency graph, where $p_i$ denotes the $i$ -th path and $v_i$ denotes the $i$ -th vertex^[21]. (a) Example of a directed graph divided into four paths. (b) Diagram of dependency relationships between paths. (c) Path dependency graph.
GraSU architecture^[22].
Forward and backward sweeping mode. Forward and backward propagation follows the order of vertice $h_i$ in the partition^[23].
Architecture of FNNG^[24].
XuLin architecture^[25]. HPU: hypergraph pocessing unit, PR: private register, AU: activation unit, HDRU: high-priority data reducing unit, and LDRU: low-priority data reducing unit.
Running time of XuLin^[25]. TW: trec-wt; OK: com-Orkut; DU: delicious-ut; OG: Orkut-group.
Illustrative example of a heterogeneous graph^[26]. (a) Vertices' three types. (b) Academic heterogeneous graph containing three types of vertices and three types of connections, where $a_i$ means the $i$ -th author, $p_i$ means the $i$ -th paper, and $c_i$ means the i-th conference. (c) Two defined metapaths (i.e., APA and APCPA). (d) Two metapath instances.
MetaNMP's running time^[26]. DP: DBLP; IB: IMDB; LF: LastFM; OM: OGB-MAG; OG: OAG.

Others

External link to attachment
https://rdcu.be/dJ4DZ
Compressed file
1941KB
DOCX format
Chinese Information 94KB
PDF format
812KB

Related articles

High-Impact Bug Report Identification with Imbalanced Learning Strategies

2017, 32(1): 181-198. DOI: 10.1007/s11390-017-1713-3
BACH:A Bandwidth-Aware Hybrid Cache Hierarchy Design with Nonvolatile Memories

2016, 31(1): 20-35. DOI: 10.1007/s11390-016-1609-7
Structure-Aware Nonlocal Optimization Framework for Image Colorization

2015, 30(3): 478-488. DOI: 10.1007/s11390-015-1538-x
A Two-Step Regularization Framework for Non-Local Means

2014, 29(6): 1026-1037. DOI: 10.1007/s11390-014-1487-9
Optimizing Parallel S_n Sweeps on Unstructured Grids for Multi-Core Clusters

2013, 28(4): 657-670. DOI: 10.1007/s11390-013-1366-9
Decomposing a Kind of Weakly Invertible Finite Automata with Delay 2

2003, 18(3)
Structure of Weakly Invertible Semi-Input-Memory Finite Automata with Delay 2

2002, 17(6)
Structure of Weakly Invertible Semi-Input-Memory Finite Automata with Delay 1

2002, 17(4)
Constructing Finite Automata with Invertibility by transformation Method

2000, 15(1): 10-26.
On the Structure of (Weak) Inverses of an (Weakly) Invertible Finite Automaton

1986, 1(3): 92-100.