Transcript CHEP 2000 Data Acquisition & Control

CHEP 2000
Data Acquisition & Control
Session Report
Chip Watson & Pierre Vande Vyvre
Overview of Sessions
Monday: Data Acquisition Systems
(“Past”, Present, and Future)
Tuesday: Detector Controls + Triggering
Wednesday: Event Building
Thursday: Run Control & Monitoring
Plus: posters covering all of these topics
Past, Present, Future
Good representative sample of experiments
• Close to end: DELPHI
• Running: BABAR, BELLE, KLOE, NA48
• Close to start: CDF, D0, Phenix, Star
• Future:ALICE, ATLAS, CMS
• Far future: Linear Collider
Data Acquisition Systems
• Online Data for Experiments -- ODE (toolkit)
– reuse of some components from DART, CDF, D0
– “Java is the implementation language of choice” for
many aspects
• BELLE DAQ System, Status & Future
– trend towards commodity (replacing custom switch)
• Automated Data Quality Monitoring…Babar
– sw pkg, 1D, 2D comparisons to ref, analytic forms...
– a very nice OO design, probably worth re-using
Data Acquisition Systems (cont.)
• …CDF Run II Data Acq System
– ATM switch, smart sockets, run control in Java, JDBC
• Alice DAQ
– Future: build @ 25 GB/s, write @ 1.25 GB/s
• Atlas DAQ / EventFilter Prototype-1
– 400x400 switch planned
– custom PMC card w/ FPGA, VHDL synthesized logic
for data readout ==> custom still lives, but only close to
the detector
• D0 Data Acq Sys & Operational Control
– NT SMPs in L3
Detector Controls
• DIM... (publish/subscribe toolkit)
– from Delphi, reused by Babar (SMI++ transport)
• Lessons learned … from Delphi
– components are good, but integration is key
• Babar Online Detector Control
– reuse of SMI++, EPICS (encapsulated in C++
component proxies)
• Phenix Ancillary Control System
– EPICS for HV, OPC (OLE for process control - MS)
for low voltage
• SW tools @ ECAL/CMS Rome
– open source good, commercial less so
Triggering
• a simulation of the L1 trigger for the CMS muon
chamber
– OO, Objectivity
• the Atlas L2 Trigger
– large testbeds: use of commodity processors &
networks looks good
• sw trigger for a tesla experiment
– read everything (1-2 GB/s) & then filter
Event Building
• CDF
• NA48
– large testbeds: thumbs up to use of commodity
processors & networks
• a reliable broadcast for CMS event builder
– investigating ways to use commodity nets to pass event
IDs
• CMS Daq Column (testbed at H2)
• CMS testbeds -- Gigabit Ethernet & Myrinet
– future: 512x512, GE test 4x4, myrinet 8x8
– 120 MB/s/link achieved now ==> looks good for
commodity stuff!
Event Building (cont.)
• Babar Online Computing System
– 90% C++, 5% Java; careful attention to interfaces
paid off (abstract API’s)
– lesson learned:
• a fully integrated prototype would have been valuable
• dependency problems in C++ worse than expected
• Data Handling & Filter Frameworks for D0 L3
Run Control & Monitoring
•
•
•
•
•
•
•
•
•
•
•
Event Logging & Distribution for Babar Online
Babar Online DB’s
CLEO III Data Acq & Controls
Process & Data Flow Control in KLOE
Online monitoring system at KLOE
Online Monitoring & Module Maint…CDF
Star Online
Usage of ROOT in the ALICE DATE system
Scalability of Atlas DAQ/EF prototype-1
Java Run Control System for CMS small daq
Java/CORBA Control & Monitoring - Phenix
Technologies: Hardware
Processors
– Intel Pentium 9
– Power PC 4
– Sparc 3
Busses & Crates
– VME 12
– PCI/PMC 4
– MXI 1
Networking
–
–
–
–
–
–
ATM 2
FDDI 1
Fibre Channel 1
Myrinet 1
Switched Ethernet 8
SCI 1
Technologies : System Software
Operating systems
–
–
–
–
–
–
–
AIX 1
Linux 9
VxWorks 4
Lynx/OS 4
OS/9 1
Solaris 3
WNT 3
Languages
–
–
–
–
–
C/C++
Java
Tcl/Tk
Python
XML
Protocols
– TCP/IP/UDP
5
– Corba 4
– COM 1
– Proprietary:1
Technologies: Application
Software
Online physics analysis
package
– IDL 1
– JAS 2
– ROOT 4
Other packages
– Objectivity
– Oracle
– EPICS
Commercial software
DBMS, middleware, GUIs
• Yes, but
– HEP is a small market.Soft will not be adapted. Has to
fit.
– Varying levels of support Sometimes below
expectation.
– Response-time to bugs too long.
– Risk to become company-dependent.
• Open-Source could help for some of these issues
Past, Present, Future
Moving to bigger systems:
– Scalability not always obvious
– Close to end: Smooth integration has
sometimes taken few years
– Close to start: “Real-size proto would have
helped.”
– Future exp: Building prototypes bigger and
bigger.
Misc. Observations
• No Controversy this year…
• HEP scaling laws . . . constants of DAQ
– minutes to fill a commodity disk
– DAQ rate compared to large commodity switch
– number of tape drives written simultaneously
=> bandwidth not the problem, complexity is
Summary: Trends
•
•
•
•
•
•
Linux
commodity PC’s, switched networks
OO, C++ solidly in place, Java emerging
re-use of components
offline software moving upstream
CORBA