Server-assisted Latency Management for Wide-area Distributed Systems

Recently many Internet services employ wide-area platforms to improve the end-user experience in the WAN. To maintain close control over their remote nodes, the wide-area systems require low-latency dissemination of new updates for system configurations, customer requirements, and task lists at runtime. However, we observe that existing data transfer systems focus on resource efficiency for open client populations, rather than focusing on completion latency for a known set of nodes. In examining this problem, we find that optimizing for latency produces strategies radically different from existing systems, and can dramatically reduce latency across a wide range of scenarios. This paper presents a latency-sensitive file transfer system, Lsync that can be used as synchronization building block for wide-area systems where latency matters. Lsync performs novel node selection, scheduling, and adaptive policy switching that dynamically chooses the best strategy using information available at runtime. Our evaluation results from a PlanetLab deployment show that Lsync outperforms a wide variety of data transfer systems and achieves significantly higher synchronization ratio even under frequent file updates.