Slide 8 of 24
Notes:
Require a distributed object store that scales to many PetaBytes (10^15 Bytes)
RDBMS wont cut it: think of a virtual table with 10^9 rows and many many columns - just for events accumulated in ~1 year of LHC operation
Need replication of large fractions of the objects to collaborating institutes:
Achievable transparently across the network or when using freight of tapes/DVDs
Object store must keep track of location and access mathod for all objects in the networked, distributed system
The user shouldn’t be concerned with the location of objects in the system
Need to access the data from a large variety of software applications
Need several language bindings
Need support for major operating systems (NT and Unix odours)
Need standards conformance(safely predict C++ -> Java -> ?? -> ?? before 2025)
Probably cannot afford the manpower to develop and maintain our own ODBMS.
