Airat A. Sadreev writes: > Why not use relational DBMS instead of ROOT I/O System? It supports > access to subset of attributes, pretty sorting, selecting, simultanious > access, etc. There are free RDBMS: Interbase, Oracle, MySQL for FreeBSD, > Linux. > Airat: the issue you brought up is too big for the answer to be simply "yes" or "no". HEP community has discussed it many times at different levels. Briefly: ROOT I/O system is designed primarily for event I/O. The most important design requirements for event I/O in HEP are different from those put into design of the relational and also object-oriented DB. The key issues for us are performance, anability to process huge (in Petabyte scale) volumes of the data and convenience for analysis. ROOT I/O inherits a lot of features polished by decades of our experience. The most important feature of database software which is missing in ROOT I/O is the support of write locks which however are not necessary for event I/O. Data handling systems used in HEP have several components with file I/O tools (what you refer to as to "ROOT I/O System") being are just one of them. The other components are the Tape Robotic System, Data Catalog, Staging System etc. The advantages of using a relational or object-oriented DB for event I/O in HEP are yet to be demonstrated. This doesn't mean however that there is no place for R(OO)DBMS in HEP at all. Online calibration databases do require multiwrite real-time access and DBMS software provide the necesary tools for it. For example, CDF experiment at FNAL will be using Oracle for online calibration DB and for the data catalog. Oracle, BTW, is very far from being free... best, Pasha
This archive was generated by hypermail 2b29 : Tue Jan 04 2000 - 00:43:41 MET