A Data-Driven Approach to Dataflow-Aware Online Scheduling for Graph Neural Network Inference

Graph Neural Networks (GNNs) have shown significant promise in various domains, such as recommendation systems, bioinformatics, and network analysis. However, the irregularity of graph data poses unique challenges for efficient computation, leading to the development of specialized GNN accelerator architectures that surpass traditional CPU and GPU performance. Despite this, the structural diversity of input graphs results in varying performance across different GNN accelerators, depending on their dataflows. This variability in performance due to differing dataflows and graph properties remains largely unexplored, limiting the adaptability of GNN accelerators. To address this, we propose a data-driven framework for dataflow-aware latency prediction in GNN inference. Our approach involves training regressors to predict the latency of executing specific graphs on particular dataflows, using simulations on synthetic graphs. Experimental results indicate that our regressors can predict the optimal dataflow for a given graph with up to 91.28% accuracy and a Mean Absolute Percentage Error (MAPE) of 3.78%. Additionally, we introduce an online scheduling algorithm that uses these regressors to enhance scheduling decisions. Our experiments demonstrate that this algorithm achieves up to $3.17\times$ speedup in mean completion time and $6.26\times$ speedup in mean execution time compared to the best feasible baseline across all datasets.

ProGNNosis: A Data-driven Model to Predict GNN Computation Time Using Graph Metrics

Graph Neural Networks (GNN) show great promise in problems dealing with graph-structured data. One of the unique points of GNNs is their flexibility to adapt to multiple problems, which not only leads to wide applicability, but also poses important challenges when finding the best model or acceleration technique for a particular problem. An example of such challenges resides in the fact that the accuracy or effectiveness of a GNN model or acceleration technique generally depends on the structure of the underlying graph. In this paper, in an attempt to address the problem of graph-dependent acceleration, we propose ProGNNosis, a data-driven model that can predict the GNN training time of a given GNN model running over a graph of arbitrary characteristics by inspecting the input graph metrics. Such prediction is made based on a regression that was previously trained offline using a diverse synthetic graph dataset. In practice, our method allows making informed decisions on which design to use for a specific problem. In the paper, the methodology to build ProGNNosis is defined and applied for a specific use case, where it helps to decide which graph representation is better. Our results show that ProGNNosis helps achieve an average speedup of 1.22X over randomly selecting a graph representation in multiple widely used GNN models such as GCN, GIN, GAT, or GraphSAGE.

Bias Reduction via Cooperative Bargaining in Synthetic Graph Dataset Generation

In general, to draw robust conclusions from a dataset, all the analyzed population must be represented on said dataset. Having a dataset that does not fulfill this condition normally leads to selection bias. Additionally, graphs have been used to model a wide variety of problems. Although synthetic graphs can be used to augment available real graph datasets to overcome selection bias, the generation of unbiased synthetic datasets is complex with current tools. In this work, we propose a method to find a synthetic graph dataset that has an even representation of graphs with different metrics. The resulting dataset can then be used, among others, for benchmarking graph processing techniques as the accuracy of different Graph Neural Network (GNN) models or the speedups obtained by different graph processing acceleration frameworks.