It is challenging to model the performance of distributed graph computation. Explicit formulation cannot easily capture the diversified factors and complex interactions in the system. Statistical learning methods require a large number of training samples to generate an accurate prediction model. However, it is time-consuming to run the required graph computation tests to obtain the training samples. In this paper, we propose TransGPerf, a transfer learning based solution that can exploit prior knowledge from a source scenario and utilize a manageable amount of training data for modeling the performance of a target graph computation scenario. Experimental results show that our proposed method is capable of generating accurate models for a wide range of graph computation tasks on PowerGraph and GraphX. It outperforms transfer learning methods proposed for other applications in the literature.

As real-world graphs are often evolving over time, interest in analyzing the temporal behavior of graphs has grown. Herein, we propose Auxo, a novel temporal graph management system to support temporal graph analysis. It supports both efficient global and local queries with low space overhead. Auxo organizes temporal graph data in spatio-temporal chunks. A chunk spans a particular time interval and covers a set of vertices in a graph. We propose chunk layout and chunk splitting designs to achieve the desired efficiency and the abovementioned goals. First, by carefully choosing the time split policy, Auxo achieves linear complexity in both space usage and query time. Second, graph splitting further improves the worst-case query time, and reduces the performance variance introduced by splitting operations. Third, Auxo optimizes the data layout inside chunks, thereby significantly improving the performance of traverse-based graph queries. Experimental evaluation showed that Auxo achieved $\mathrm{2.9}\mathrm{\times }$ to $\mathrm{12.1}\mathrm{\times }$ improvement for global queries, and $\mathrm{1.7}\mathrm{\times }$ to $\mathrm{2.7}\mathrm{\times }$ improvement for local queries, as compared with state-of-the-art open-source solutions.