Optimizing parallel access to the BaBar database system using CORBA servers

The BaBar Experiment collected around 20 TB of data during its first 6 months of running. Now, after 18 months, data size exceeds 300 TB, and according to prognosis, it is a small fraction of the size of data coming in the next few months. In order to keep up with the data, significant effort was put into tuning the database system. It led to great performance improvements, as well as to inevitable system expansion-450 simultaneous processing nodes alone used for data reconstruction. It is believed that further growth beyond 600 nodes will happen soon. In such an environment, many complex operations are executed simultaneously on hundreds of machines, putting a huge load on data servers and increasing network traffic. Introducing two CORBA servers halved startup time, and dramatically off loaded database servers: data servers as well as lock servers. The authors describes details of design and implementation of two servers recently introduced in the BaBar system: Conditions OID Server and Clustering Server. The first experience of using these servers is discussed. A discussion on a Collection Server for data analysis, Currently being designed is included