Cecile Le Pape (Laboratoire d'Informatique de Paris 6 (LIP6))
Stephane Gancarski (Laboratoire d'Informatique de Paris 6 (LIP6)) Relaxing replica freshness has been exploited in database clusters to optimize load balancing. However, in most approaches, refreshment is typically coupled with other functions such as routing or scheduling, which makes it hard to analyze the impact of the refresh strategy itself on performance. In this paper, we propose to support routing-independent refresh strategies in a database cluster with mono-master lazy replication. First, we propose a model for capturing existing refresh strategies. Second, we describe the support of this model in Refresco, a middleware prototype for freshness-aware routing in database clusters. Third, we describe an experimental validation to test some typical strategies against different workloads. The results show that the choice of the best strategy depends not only on the workload, but also on the conflict rate between transactions and queries and on the freshness level required by queries. Although there is no strategy that is best in all cases, we found that one strategy is usually good and could be used as default strategy. This work was partially financed through the ANR-ARA Respire project. Keywords: replication, database cluster, load balancing, refresh strategy. Replication and fault-tolerance in Grids , Indexing, caching and load balancing in Grids , Grid performance evaluation

Download the fullpaper