Databases in ALICE

1
Databases in ALICE

• Fons Rademakers

2
ALICE in Numbers (est.)

• Raw event size 25 MB (1 MB p-p)
• Trigger rate 50 Hz
• Rate to HSM 1.25 GB/s
• Total stored per year 3-4 PB/y

3
Important Decisions Taken

• Decided to use ROOT as framework
• Decided to use ROOT as data store for all write
  once, read often data
• RAW
• ESD
• AOD
• TAG
• Decided to use RDBMS for all other data
• File catalogues, calibration data, etc.

4
Why?

• ROOT is a robust framework used by a large
  community, of which many are working on identical
  problems
• Using a storage technology we own and fully
  understand for the bulk of our data is essential
• Also performance, efficiency, scalability, etc.
• Standard SQL based RDBMS systems are a good
  solution for typical relational data

5
ALICE Data Challenges

• This hybrid solution has been tested in
  practice during the first 3 ADCs
• ADC 1 30 MB/s peak, 7 TB to HPSS
• ADC 2 100 MB/s peak, 23 TB to HPSS/CASTOR
• ADC 3 120 MB/s peak, 110 TB to CASTOR
• ROOT for event store, MySQL for file catalogue

6
ALICE Data challenges
CASTOR
GRID
7
ALICE ADC III
Event building (DATE)
Recording (ROOTRFIO)
Mass Storage (CASTOR)
8
DATE ROOT I/O
9
DATE ROOT I/O CASTOR
Writing to local disk
Migration to tape
10
DATE ROOT I/O CASTOR
11
ADC III Highlights

• Excellent system stability during 3 months
• DATE throughput 550 MB/s (max) 350 MB/s
  (ALICE-like)
• DATEROOT throughput 240 MB/s
• DATEROOTCASTOR throughput 120 MB/s
• average during several days 85 MB/s (gt 50
  TB/week)
• 2200 runs, 2 107 events, longest DATE run (86
  hours, 54 TB)
• 500 TB in DAQ, 200 TB in DAQROOT I/O
• CASTOR gt 100.000 files of 1 GByte gt 110 TBytes
• File catalogue with 105 entries

12
ADC Plans
13
Simulation Production

• The AliROOT simulation and reconstruction program
  produces ROOT files
• Massive distributed production is planned mid
  next year for Physics Performance Report
• Bookkeeping in central RDBMS (MySQL/Oracle)
• Job run status
• Job history
• Distributed file catalogues

14
ALICE and the GRID

• Active player in WP8
• ALICE long term project is distributed
  interactive data analysis using PROOF
• Will interface to the GRID via ROOT

15
ALICE and DataGrid
Local
Remote
16
DataGrid ROOT
Root RDB
Selection parameters
Grid RB
Spawn PROOF tasks
Update Root RDB
17
ALICE and Oracle 9i

• We are interested in evaluating Oracle for those
  areas where we now use RDBMSs
• Access can be via ROOTs RDBC module, which is
  JDBC 2.0 compliant and allows object-relational
  access to any RDBMS supporting an ODBC driver