Lightweight Fault Tolerance in Large-Scale Distributed Graph Processing

The success of Google’s Pregel framework in distributed graph processing has inspired a surging interest in developing Pregel-like platforms featuring a user-friendly “think like a vertex” programming model. Existing Pregel-like systems support a fault tolerance mechanism called checkpointing, which periodically saves computation states as checkpoints to HDFS, so that when a failure happens, computation rolls back to the latest checkpoint. However, a checkpoint in existing systems stores a huge amount of data, including vertex states, edges, and messages sent by vertices, which significantly degrades the failure-free performance. Moreover, the high checkpointing cost prevents frequent checkpointing, and thus recovery has to replay all the computations from a state checkpointed some time ago. In this paper, we propose a novel checkpointing approach which only stores vertex states and incremental edge updates to HDFS as a lightweight checkpoint (LWCP), so that writing an LWCP is typically tens of times faster than writing a conventional checkpoint. To recover from the latest LWCP, messages are generated from the vertex states, and graph topology is recovered by replaying incremental edge updates. We show how to realize lightweight checkpointing with minor modifications of the vertex-centric programming interface. We also apply the same idea to a recently-proposed log-based approach for fast recovery, to make it work efficiently in practice by significantly reducing the cost of garbage collection of logs. Extensive experiments on large real graphs verified the effectiveness of LWCP in improving both failure-free performance and the performance of recovery.