Toward Fast and Reliable Active-Active Geo-Replication for a Distributed Data Caching Service in the Mobile Cloud
Daniel House, Heng Kuang, Kajaruban Surendran, Paul Chen · Procedia Computer Science · 2021
In this paper, we describe our experiences extending a distributed data caching service in the mobile cloud to support active-active geo-replication. We show how our enhancements guarantee data consistency between regions after a network partition recovery. An approach is presented for keeping geo-distributed replicas synchronized despite the cache data operations replication not ensuring causal delivery in the presence of long network partitions. We use Redis, one of the most popular in-memory databases for the distributed data caching service in the mobile cloud, as a proof of concept to apply our approach in a plug-in way (minimizing the impact on both the server and client side of the cache service). Redis exposes a powerful extension API that allows new abstract data types to be associated with keys but does not provide direct support for adding and managing global dictionary metadata, which we added in our solution. That extension API is used to add the CRDT (Conflict-free Replicated Data Type) to resolve the writing conflicts from multiple regions.