The Grid-distributed data analysis in CMS

The CMS experiment will soon produce a huge amount of data (a few P Bytes per year) that will be distributed and stored in many computing centres spread across the countries participating in the collaboration. Data will be available to the whole CMS physicists: this will be possible thanks to the services provided by supported Grids. Crab is the CMS collaboration tool developed to allow physicists to access and analyze data stored over world-wide sites. It aims to simplify the data discovery process and the jobs creation, execution and monitoring tasks hiding the details related both to Grid infrastructures and CMS analysis framework. We will discuss the recent evolution of this tool from its standalone version up to the client-server architecture adopted for particularly challenging workload volumes and we will report the usage statistics collected from the Crab community, involving so far almost 600 distinct users.