Cern plans to store between several hundred terabytes and a few petabytes of data. Over 10 years, the database would need to store 100Pbytes, said Jamie Shiers, database group leader at Cern.
An Oracle customer for 20 years, Cern has rolled out the technology throughout the organisation, using it to book meeting rooms, manage the network and power e-business operations.
The organisation has transferred 300Tbytes of data from an object database to Linux to Oracle since 2002. It is using Oracle databases and application server software to track this data and to schedule computational work.
By early 2004, Cern plans to use the Real Application Clusters feature to provide high levels of availability on Cern's grid.
"If the application related to the grid is unavailable, the grid halts and you could lose all the computational work queued up to run over the grid," said Shiers.
Shiers also highlighted the improvements Oracle has made in processing "native numbers", which are used heavily in the scientific community. "The Oracle database was always efficient when dealing with figures such as $99.99, but it did not perform when handling calculations," he said.
Shiers would like those numbers to be stored in the database with no overheads, but if 100Pbytes of data were stored, an overhead with a factor of two would double the storage requirements.
The Oracle database has also been used to decommission the previous particle accelerator. "Every single component, every magnet, every vacuum tube and every nut and bolt needs to be logged in the database," he said. Moreover, Shiers said the components need to be logged permanently. "We are not sure how to do this," he said. "It is quite a demanding application."