Excavating the Potential of Graph Workload on RDMA-based Far Memory Architecture

Disaggregated architecture brings new opportunities to memory -consuming applications like graph processing. It allows one to outspread memory access pressure from local to far memory, providing an attractive alternative to disk-based processing. Although existing works on general-purpose far mem-ory platforms show great potentials for application expansion, it is unclear how graph processing applications could benefit from disaggregated architecture, and how different optimization methods influence the overall performance. In this paper, we take the first step to analyze the impact of graph processing workload on disaggregated architecture by extending the GridGraph framework on top of the RDMA-based far memory system. We design Fargraph, a far memory coordi-nation strategy for enhancing graph processing workload. Specif-ically, Fargraph reduces the overall data movement through a well-crafted, graph-aware data segment offloading mechanism. In addition, we use optimal data segment splitting and asynchronous data buffering to achieve graph iteration-friendly far memory access. We show that Fargraph achieves near-oracle performance for typical in-local-memory graph processing systems. Fargraph shows up to 8.3 x speedup compared to Fastswap, the state-of-the-art, general-purpose far memory platform.

I. Introduction

Today's various graph applications demand better memory performance at different graph scales [18], [32], [36], [39], [40]. In the past, most graph applications can be processed by a single-node system given the relatively small size of the graph in existing in-memory graph frameworks [29], [39], [40]. Distributed graph frameworks are required only for very large-scale data analytic problems due to the communication overhead [24], [35], [41]. Nevertheless, as shown in Figure 1-(a), many graph frameworks mainly focus on medium-sized graphs (from 1GB to several hundreds of GB) [17], [36], [42]. Although current out-of-core graph computing frameworks could handle medium-sized graphs with external storage, they suffer performance degradation due to the I/O bottleneck.