Users often need to integrate large amounts of RDF data from multiple sources. Although there has been a lot of work on data cleaning and integration for structured data, relatively little work has been done for RDF data. We consider two different approaches to data integration: 1) a traditional data warehouse approach where relevant RDF data is extracted from different sources and then integrated in a data warehouse; 2) a virtual integration approach where RDF data still resides at each source and data integration happens when the data is queried, through a mediator that coordinates with wrappers at each source. It is often unclear how to choose the appropriate approach given an application scenario. This paper proposes RDFINT, a benchmark to compare these two approaches for integrating RDF data. We describe typical data integration operations, metrics that can be used to compare these two approaches, and factors that affect these metrics. We also report preliminary results of an implementation of these two approaches using the Apache Jena Fuseki framework.