SiHGNN: Leveraging Properties of Semantic Graphs for Efficient HGNN Acceleration

TL;DR

The paper tackles inefficiencies in accelerating HGNNs on GPUs caused by irregular execution and data movement during semantic graph generation and feature processing. It introduces SiHGNN, a lightweight accelerator frontend comprising a tree-based Semantic Graph Builder and a Graph Restructurer to exploit semantic graph properties. The Semantic Graph Builder uses a Callback Trie Tree to reuse sub-metapaths and avoid redundant computation, while the Graph Restructurer partitions graphs into subgraphs with strong community structure to reduce buffer thrashing. Experimental results show SiHGNN achieving an average speedup of 2.95x over the state-of-the-art HGNN accelerator, with modest area and power overhead, demonstrating practical benefits for inductive HGNN inference.

Abstract

Heterogeneous Graph Neural Networks (HGNNs) have expanded graph representation learning to heterogeneous graph fields. Recent studies have demonstrated their superior performance across various applications, including medical analysis and recommendation systems, often surpassing existing methods. However, GPUs often experience inefficiencies when executing HGNNs due to their unique and complex execution patterns. Compared to traditional Graph Neural Networks, these patterns further exacerbate irregularities in memory access. To tackle these challenges, recent studies have focused on developing domain-specific accelerators for HGNNs. Nonetheless, most of these efforts have concentrated on optimizing the datapath or scheduling data accesses, while largely overlooking the potential benefits that could be gained from leveraging the inherent properties of the semantic graph, such as its topology, layout, and generation. In this work, we focus on leveraging the properties of semantic graphs to enhance HGNN performance. First, we analyze the Semantic Graph Build (SGB) stage and identify significant opportunities for data reuse during semantic graph generation. Next, we uncover the phenomenon of buffer thrashing during the Graph Feature Processing (GFP) stage, revealing potential optimization opportunities in semantic graph layout. Furthermore, we propose a lightweight hardware accelerator frontend for HGNNs, called SiHGNN. This accelerator frontend incorporates a tree-based Semantic Graph Builder for efficient semantic graph generation and features a novel Graph Restructurer for optimizing semantic graph layouts. Experimental results show that SiHGNN enables the state-of-the-art HGNN accelerator to achieve an average performance improvement of 2.95$\times$.

Figures (18)

• Figure 1: An example of HetGs and execution process of HGNN models.
• Figure 2: The number of semantic graphs and the normalized time of the SGB stage across various lengths of the metapaths.
• Figure 3: Analysis on T4 GPU with various models: The L1 and L2 cache hit rate during NA sub-stage.
• Figure 4: Analysis of HiHGNN with the RGCN model: Frequency of vertex feature replacements during the NA sub-stage. The datasets, from left to right, are ACM, DBLP, and IMDB.
• Figure 5: Design and workflow overview of SiHGNN.