The Chandra Data Archive is distributed at three physically remote locations, two of them in Cambridge, MA and a third in Leicester, UK. Each installation operates local hardware and a locally configured software release. The data are stored at a single location or in synchronized copies at multiple locations. The architecture enables processes to access the installation that is closest to the user or another installation if the first becomes overloaded or unavailable. This paper presents the archive architecture for the multiple installations. We explain the mechanisms that synchronize the data and we analyze the differences in data holdings across sites. We discuss how the software release is configured to operate at each installation and how users are routed to an installation depending on their profile. Finally, we describe the load balancing and failover mechanisms built into the archive.
Astronomical Data Analysis Software and Systems (ADASS) XIII
- Pub Date:
- July 2004