Sprinkler - Reliable Broadcast for Geographically Dispersed Datacenters

This paper describes and evaluates Sprinkler, a reliable highthroughput broadcast facility for geographically dispersed datacenters. For scaling cloud services, datacenters use caching throughout their infrastructure. Sprinkler can be used to broadcast update events that invalidate cache entries. The number of recipients can scale to many thousands in such scenarios. The Sprinkler infrastructure consists of two layers: one layer to disseminate events among datacenters, and a second layer to disseminate events among machines within a datacenter. A novel garbage collection interface is introduced to save storage space and network bandwidth. The first layer is evaluated using an implementation deployed on Emulab. For the second layer, involving thousands of nodes, we use a discrete event simulation. The effect of garbage collection is analyzed using simulation. The evaluation shows that Sprinkler can disseminate millions of events per second throughout a large cloud infrastructure, and garbage collection is effective in workloads like cache invalidation.