Presentation is loading. Please wait.

Presentation is loading. Please wait.

BTeV Database Workshop Summary1 BTeV Database Workshop Summary D. Menasce - I.N.F.N. Milano Day 1 1. Database design database design (constraints) application.

Similar presentations


Presentation on theme: "BTeV Database Workshop Summary1 BTeV Database Workshop Summary D. Menasce - I.N.F.N. Milano Day 1 1. Database design database design (constraints) application."— Presentation transcript:

1 BTeV Database Workshop Summary1 BTeV Database Workshop Summary D. Menasce - I.N.F.N. Milano Day 1 1. Database design database design (constraints) application considerations architectures (3 tier vs 2 tier) when to use what database middleware - and interfaces process of the DB design - what to start from DB designer - how to become a specialist 2. Comparison of databases and tools possible contenders and their estimated longevity associated design tools Arguments of discussion

2 BTeV Database Workshop Summary2 Day 2 1.Run II experiences What worked and didn't. What should/should not be in databases. Day 3 1.DB applications and requirements What kind of DB applications we can foresee? What can be a list of an application requirements? purpose of the database source of the data -- where is the information that populates the database generated? quantity of the data anticipated use of the data/access patterns

3 BTeV Database Workshop Summary3 1. Database design Important concern is evolution schema in the design On-line and off-line DB have different requirements DB #1: On-line Database –Maintained at the experiment –Critical for data taking –Small (100 GB), Low Rate (a few transactions/s). DB #2: Off-line Database –Maintained in Feynman Computing Center –Critical for data processing and analysis –Large (100 ’ s of GB), High rate (several 10 ’ s of transactions/s), grows 1 GB per day. (Lueking, D0) Keep in mind Joel ’ s Prime Directive BTeV will achieve its physics goals architectures

4 BTeV Database Workshop Summary4 CDF experience shows Three Tiers of Management is a good approach PhysicistsOraclesDBA Applications Programmers Always remember sociology is an important constrain: DBA: they speak SQL Oracles: they speak both C++ and SQL AP: they speak C++ and even order breakfast in SQL Physicists: Speak E=mc 2 and make dog ’ s dinner of C++ Provide excellent public interfaces and documentation (St.Denis, CDF) Design should conceal horrendous details (Corba backends and such) from final user architectures

5 BTeV Database Workshop Summary5 when to use what database MySQL or PostgreSQL Is ORACLE a competitor ? (Semenov) Continental divide is difficult to define, but criteria are:  Time critical (real time?)  Size  Data integrity  Back ups  Very advanced features ORACLE for heavy-duty (proven efficiency) but expensive. Requires big effort in terms of number of experts. MySQL and/or Postgres good candidates for middle size DB Comparison of databases and tools

6 BTeV Database Workshop Summary6 MySQL vs PostrgreSQL performance Concurrency test results Mysql dies9.48 page/s90 clients Mysql dies9.28 page/s120 clients 15.43 page/s10 page/s50 clients 16.03 page/s10.25 page/s30 clients MySQLPostgreSQL (Semenov)

7 BTeV Database Workshop Summary7 MySQL vs PostrgeSQL No, plan v4.0yesForeign keys NoyesUser def types No, plan v4.1yessubselect No, plan v4.1yesStored procedures No, plan v4.2yesViews No, plan v4.1yestriggers No, plan v4.2yesconstraints Yes since v3.23yestransactions MySQLPostgreSQL (Semenov)

8 BTeV Database Workshop Summary8 process of the DB design - what to start from DB Roll-Out Procedure Rough draft. Plan with rough idea of what the functionality, data and relationships among the tables might be. Use Oracle Designer to create the schema design, and designer repository. Employ the DDL which is created from Designer to create the tables, keys, triggers, etc. All development is performed using a "development" instance of the database. DDL saved in cvs. Must provide rate and sizing information for DBA's. An INT instance is created with controlled DDL and thoroughly tested. When the application is ready for production use, DDL has version control imposed and the application is transitioned to PRD database instance. The methodology for version cutting and evolution is written up. Design by experts with physicists sitting on their shoulders Many iterations to converge to desired functionality (evolution schema again) (Lueking, D0) Provide testing tools at early stage (both HW platforms and SF components)

9 BTeV Database Workshop Summary9 DB designer - how to become a specialist Tier schema helps: no need to become a real expert (maybe) Having so many database developers is complex. The middle tier architecture added significantly to the development time. This was exacerbated (to say the least) by the middle tier tools not being ready in time. Probably worth it, for most applications. Having the databases administered centrally by professionals is a real good idea. (Lueking, D0) Central problem in overall BTeV software Should we address problem of transition to modern technologies organizing “ schools ” with experts? (C++, OO issues, DB issues, …)

10 BTeV Database Workshop Summary10 1.Run II experiences What worked and didn't. What should/should not be in databases. Online Calibration Coordinator Sets policy for calibration writers Liaison for support software First line debug Provides a team that can carry a pager CDF Does not Have one! No major failure/problem reported by D0 and CDF Big effort by D0 to provide fail-over service: no need at the end since system (SUNs) proved to be very reliable (can we afford same approach?) (St.Denis, CDF)

11 BTeV Database Workshop Summary11 DB applications and requirements 1) Detector construction tracking databases: 2) Calibration databases: 3) Configuration databases: 4) Monitoring databases: 5) Event-related databases: 6) Analysis and Simulation support databases: 7) Documentation and general information databases: For each envisaged category, a WEB interface is considered essential APIs of database to scripting languages is important to this extent Design of WEB server is crucial for efficiency (application server better that CGI) On-line and off-line applications have crossed access paths to almost all these DB categories. Again, good design in the de-coupling of the core services and the application front-end is crucial. For certainty some category has escaped this zero-th order list: good insulation layer could provide room for later addition with minimal impact (hopefully). Longevity of the DB is an overall requirement: how do we access data after 10 years? What are the tools of the trade to accomplish this? Answers were not given in this workshop.


Download ppt "BTeV Database Workshop Summary1 BTeV Database Workshop Summary D. Menasce - I.N.F.N. Milano Day 1 1. Database design database design (constraints) application."

Similar presentations


Ads by Google