高效的图数据处理如今备受关注，广泛应用于众多领域如社交网络，电子商务，生物信息，金融贸易，以及欺诈检测等。设计图数据应用感知系统架构快速成为重要的研究方向，对于促进关联型大数据的高效及时分析具有赋能意义。然而，现有的工作主要关注于处理一个单一的大图，对于并发处理多个中小规模的图计算请求则缺乏研究和探索。在这个项目中我们率先探索一个新的计算模式：异构架构上的多程序图计算。具体来说，我们聚焦于以下三个关键问题：1）如何定义和建模多程序图负载的运行时行为，2）如何在异构体系结构上充分释放多程序图负载的性能潜力，3）如何借助近似计算技术来提升多程序图计算的可扩展性。在这个项目中，我们将同时从理论和实践两方面研发新的基本方法，以解决上述挑战。本项目将提出一套跨层协同设计的技术框架，以及一系列软硬件的自适应优化方法，相关成果将为图计算系统和图处理应用领域打开新的一扇大门。

Efficient graph processing has attracted significant attentions due to its wide applications in various domains such as social network, e-commerce, bioinformatics, finance, trading, and fraud detection, etc. Designing graph application-aware system architecture has emerged as the key enabler for processing highly interconnected big data in an efficient and timely fashion. Existing research, however, primarily focuses on analyzing a single, large-scale graph data-set, leaving the efficient processing of multiple mid-sized graph analytic queries an intriguing yet challenging open problem. In this research, we take the first to investigate a new computing paradigm: multi-programmed graph computing on heterogeneous architectures. Specifically, we focus our attention on the following key scientific problems: 1) how to define and model the run-time behavior of multi-programmed graph workload; 2) how to unleash the potential of heterogeneous architecture to boost the performance of multi-programmed graph workload; 3) how to leverage approximate computing techniques to greatly improve the scalability of multi-programmed graph computing. In this project, new, fundamental methods will be explored in both theoretical analysis and practical implementations to address the above issues. The proposed research features a cross-layer co-designed framework and various adaptive optimization systems of software and hardware. The results of this project will open the door for a new class of graph processing and analytics.