WYT: Optimized Consistency for Geo-Diverse Online Social Networks

Large scale Online Social Networks (OSNs) like Facebook and Twitter are hosted out of multiple geo-diverse data centers to provide low latency and high fault tolerance. Such geo-diversity creates large amounts of WAN traffic for maintaining the consistency of replicas at different locations. Despite the dropping price of WAN bandwidth, the growth rate of OSNs combined with the incorporation of media rich long tail content (including images and videos) makes WAN traffic costs an increasing concern for OSN operators. At the heart of the problem lies a tradeoff between consistency and WAN bandwidth cost. In this paper, we propose the "Wait Your Turn"; WYT system that optimizes the tradeoff by leveraging: (i) knowledge of mapping between social relationships and geographic location, and (ii) knowledge of timing regularities in end user activity patterns. We quantify the benefits of such an OSN-aware update propagation strategy through a trace-driven analysis and show that it reduces WAN traffic by 55% compared to an immediate update of all replicas, while having minimal impact on consistency. Furthermore, for a given budget for WAN bandwidth, WYT increases consistency by several orders of magnitude compared to FIFO scheduling of updates.