The%20story%20of%20BaBar:%20an%20IT%20perspective

1
The story of BaBar an IT perspective

• Roger Barlow
• DESY 4th September 2002

2
Example ????? ?- ?? ?0???

• BaBar
• (Work in progress)

• OPAL

3
History
1995 TDR C
1999 First data 0.5 fb-1 Objectivity / kanga
2000 First results 23 fb-1 File transfer
2001 Sin 2b 0.59?0.14?0.05 39 fb-1 Tier A sites
2002 Sin 2b 0.741?0.067?0.029 28 fb-1 Grid
4
DatatakingDrinking from a firehose
4.2 ns
32 SUN U5
L1 Trigger
L3 Trigger
1-2kHz
200 CPUs
OPR farm
PowerPCs in VME crates
Data store
30-100 Hz
200 kB/evt
5
C

• Decision taken approved languages C and F90
• (Nobody used F90)
• HEP escapes at last from FORTRAN!
• Other experiments follow
• Software immediately becomes VERY object-oriented

Unorthodox but very nice introduction to
programming Amazon.com review
6
C in practice

• Design
• Old New

Abstract objects EmcClusterAbsRecoCalo AbsRecoCa
lo AbsEvtObj EmcClusterMoments
EmcAbsClusterProperty EmcClusterDistance
EmcAbsClusterProperty
The software was written on time and it worked.
7
Some lessons

• Calibration
• System impressive technically
• Cant do what you want it to

• RogueWave
• Commercial class library
• Built in to software from start
• Superseded by STL, CLHEP
• Licence arrangements restrictive/expensive
• Now (painfully) removed

Event Display Powerful system (Event display
server) written by experts Unwieldy and
unreliable for non-experts away from SLAC Little
used
8
Objectivity

• Persistent data
• Objects still present when a job is run again-
  stored on disc
• Need to handle pointers
• Need to match data on disk to objects the program
  knows about
• Need to do this fast and furiously
• Objectivity does this no alternative

• Performance issues in OPR solved (hard
  workvendor co-operation)
• Performance issues in analysis not satisfactory
  (physicists more random)? KANGA
• Long-term viability?

9
Success the penalty
PEPII delivers and BaBar can handle it Design
Luminosity 3 1033 achieved Now 5 1033 Heading
for 4 1034 BaBar efficiency 95 (This is a
race. KEK and Belle are close behind!)

• Data handling
• Need to provide
• Disk (or tape) for data storage
• Machines for OPR
• Machines for analysis
• Moores Law is not strong enough
• Need new computing model
• Central ? Distributed

10
File Transfer
Data transfer From SLAC to outside sites Large
(RAL, IN2P3) and small (university) Use network
transfer rather than shipping tapes ESNET in US,
JANET GEANT in Europe Use RAL, Rome, etc as
staging posts
File copy programs bbcp and bbftp developed, now
standard Copy with multiple streams Achieve
80-100 Mbit/s Data available within days Chasing
improvement continuous task
11
Spreading the load

• Serious computing power at outside institutes

Each UK BaBar group 1TB fast disc
(1998) Compute farm (80 CPU) (2000) Used for
analysis Used for MC simulation
12
RAL Tier A Centre
UK GRID Funds available Tier 1/A centre at
RAL 200 CPUs 35 TB disk this year (increasing as
price drops) Rebate on common fund
contribution Everybody happy

• Similar centres at IN2P3, Bologna, plus Karlsruhe
  in future

13
Future Development

• Many users running jobs
• on distributed data
• with distributed CPUs
• Dont want technical detail

GRID provides the answer User specifies data
description and job Grid tools locate files, find
CPUs, submit jobs BaBar today LHC tomorrow
MCRB
14
Authentication
Certificate Authorities Certificates issued by
trusted organisations to people they know and
trust (or people they trust know and
trust) Establish mutual trust within BaBar for
UKHEP,CNRS, etc But authentication is not
authorisation!
Grid Certificate
Uses RSA encryption to prove to other computers
(gatekeepers) that you are who you say you are
15
Authorisation
SLAC
MANCHESTER
BaBarGrid sites
gridmap files
Grid Certificate
BaBar VO list
cron job
cron job

• Users with Grid certificates authorised to use
  resources at all BaBarGrid sites
• Getting on this list is made easy (for BaBar
  users with a Grid certificate)
• Dynamic account pool avoids need for 500x50
  userids

16
The BaBar VO
17
The BaBar Grid Demonstrator

• Rudimentary Grid submission and retrieval
• Full features to be added later

18
The Future Some Possibilities
Join EDG testbed rollout
More Compute and Storage farms
Desk-to-desk Video conferencing
Extend VO system for authorisation
SRB as universal file system
Metadata specification using xml
Framework VI