1. LHC experiments Requirements and Concepts ALICE

3. LEP in

4. … and in 2000

8. Outline ALICE general description Requirements Architecture Software
Data Challenges

9. Two running modes Dr Jekyll… … and Mr Hyde pp beam Pb-Pb collisions
general-purpose heavy ion experiment
… and Mr Hyde
pp beam
large cross-section pp processes

10. ALICE data rates 1 month (106 s) 1 Minimum Bias 20 1 - 87 Central
Dielectrons
200
Dimuon
670
24.5
2.5
1.25 NA
0.5
0.1
500 2
Pb-Pb run
pp run
Event rate (Hz)
Event size (MB)
Data in DAQ (GB/s)
Data in EB (GB/s)
Data on tape (GB/s)
10 months
Run period
Total on tape (PB)
Trigger type

11. The original architecture

12. Detector Data Link Functions: main interface with the detectors
handle detector-to-LDC data flow
handle LDC-to-detector commands & data
Keywords:
cheap
small
functional
rad-hard
long distance
optical
used everywhere

13. Local Data Concentrator
Functions:
handle and control the local DDL(s)
format the data
perform local event building
allow monitoring functions
ship events to the event builders (GDCs)
Keywords:
distributed
good data moving capabilities
from the DDL
to the Event Building Link
CPU power not indispensable
Not a farm

14. Global Data Collector Functions: accept the data sent from the LDCs
perform final event building
ship the events to the Permanent Data Storage (PDS)
Keywords:
distributed
good data moving capabilities
from the LDCs
to the PDS
CPU power not indispensable
farm

15. Event Destination Manager
Functions:
collect availability information from the GDCs
distribute event distribution policies to the data sources
Keywords:
optimized network usage
look ahead capabilities

16. Event Building Link Functions: Move data from the LDCs to the GDCs
Keywords:
big events (1-3, MB)
low rates (20, 500, 670 Hz)
many-to-many
mainly unidirectional

17. Overall key concepts Keep forward flow of data
Allow back-pressure at all levels (DDL, EBL, STL)
Standard Hw and Sw solutions sought:
ALICE collaboration
CERN computing infrastructure
Whenever possible go COTS
During the pp run, keep any unused hardware busy

18. Mismatch of rates Recent introduction of:
Transition Radiation Detector (TRD)
Dielectron trigger
change in Pixel event size
increase in estimated TPC average occupancy
Required throughput an order of magnitude too high!
New scenarios:
region-of-interest readout
online compression
online reconstruction
introduction of a level 3 trigger

19. The new architecture

20. The Event Building process
Events flow asynchronously into the LDCs
Each LDC performs - if needed - local event building
The Level 3 farm - if present - is notified
Level 3 decision - if any - is sent to LDCs and GDC
All data sources decide where to send the data according to:
directives from the Event Destination Manager
the content of the event
The chosen GDC receives:
sub-events
optional reconstructed and compressed data
optional level 3 decision
The Event Building Link does the rest

21. Software environment DATE
Data acquisition environment for ALICE and test beams
Support DDLs, LDCs, GDCs and liaison to the PDS
Standalone and complex DAQ systems
Integrated with HPSS and CASTOR (via CDR)
Keywords: C
TCP/IP
Tcl/Tk
Java
ROOT

22. Data challenges Use state-of-the-art equipment for real-life exercise
: Challenge I
7 14 MB/s, 7 TB
: Challenge II
2 * 7 max 100 MB/s, > 20 TB
transfer simulated TPC data
23 LDCs * 20 GDCs (AIX/Solaris/Linux)
with offline filtering algorithms and online objectifier (ROOT)
two different MSS (HPSS and CASTOR)
several problems  limited stability

23. Data Challenge II

28. Event building network
Pure Linux setup
20 data sources
FastEthernet local connection
GigaBit Ethernet backbone

29. Run log

30. Data Challenge III Will run during the winter 2000-2001 shutdown
Target: 100 MB/s (or more) sustained over [7..10] days
Improved stability
More “ALICE like” setup
abandon older architectures still in use at the test beams
Implement 10% of the planned ALICE EB throughput
Integrate new modules & prototypes:
improved event building
Level 3
Regional Centers
Will use the LHC computing testbed
Better status reporting tools: use PEM if available