escience and particle physics roger barlow escience showcase may 1 st 2007
TRANSCRIPT
eScience and Particle Physics
Roger Barlow
eScience showcase
May 1st 2007
Particle physics yesterday
Particle physics has pushed the computing envelope ever since the early days
- though in the early days the envelope was smaller
Particle physics today
• Today we have millions of extremely complicated events to interpret
Particle physics tomorrow
• Tomorrow the LHC will give us trillions of much more complicated events to interpret
What is our computing?
• Basic tasks fairly simple – find tracks from set of points, find particle momenta from track curvature, etc.
• Data is pretty well constrained and homogenous.
• But we need to do this on a big scale, for hundreds of millions of events.
Big is beautiful
If all the data from the LHC in a year were written to CDs, and the CDs put in a pile, that pile would be 20 km high
Commodity computing saved us in the 1990s: PCs got cheap. But that’s not enough
The Grid
Data analysis can only be achieved as a world-wide exercise using all available computing resources
This needs a Grid:
The UK GridPP project
and EGEE across Europe
What is a Grid?
Answer 1 (for journalists)
The Grid is a virtual supercomputer
Answer 2 (for politicians)
Computing coming out of a wall socket, just like electric power
Answer 3 (for the rest of us)
The Web is computers talking to computers, saying ‘Give me a certain document’.
The Grid is computers talking to computers. They can say anything.
Issues
Many computers available in computer centres
Many people who want to use them
How do the people know the computers are there? What facilities they provide
How do the computer centres know which users are allowed to do what? That users are who they say they are?
Traditional solution; every user needs an account and password at every centre
Does not scale to large numbers of computers and users
Solution – Grid certificates
• User details encoded by the RSA algorithm from a trusted Certificate Authority
• Proves that whoever presents the certificate are who they say they are. (Modern analogue of the royal seal)
• User needs only one Grid certificate – access everywhere
• Led (in academia) by UK/European particle physicists
/C=UK /O=eScience
/OU=Manchester /L=HEP
/CN=roger john barlow
UK CA
Solution - VOMS
Users join ‘Virtual Organisations’Centres negotiate with VOs for usage rights
UK VOMS system run from Manchester (Physics for GridPP, Computing for NGS). 500 users, 15 VOs, and growing
CENTREUser
VOVOMS system
Solution: pool accounts
Developed and implemented by Manchester physics
Neither user nor centre wants individual accounts Single account (‘griduser’) will lead to jobs deleting each
others’ files
Create generic accounts (ATLAS001, ATLAS002…)User assigned a generic account – can run jobs
Account is linked to certificate name so there is an audit trail for antisocial behaviour
Solution - GridSite
GridSite developed at Manchester to manage websites using grid credentials
Now used to ‘gridify’ Web Services used on the grid
GridPP website is maintained at Manchester using GridSite
Facility – Tier2
1000 dual 2.8 GHz Xeon nodes
½ Petabyte of storage10 Gbit/s networkBought as faculty investment
in eScienceReynolds House machine
roomManagement through
particle physics
Results
Working anddeliveringCPU cycles
High uptime and reliability
Who’s using it?
Manchester LHC experiments
Non-Manchester LHC experiments
Manchester non LHC experiments
Non-Manchester non-LHC experiments
Non-Particle physics users
Biomed
Drug design for combating bird flu & other diseases
ATLAS
ATLAS trigger testsLarge scale software tests cannot
be done at CERN as only ~10 computers available
Tests run at Manchester (400 CPUs) instead
ATLAS monitoring and calibrationDetectors need frequent calibrating
to convert raw signals into useful co-ordinates correctly
Computers at CERN dedicated to actual DAQ
Solution: ship raw data to Manchester. Process. Ship calibration data back
Successful large-scale tests show this is possible
Trigger
Detector
DAQ
Data storage
CAL DATA
Manchester
2000 CPUs
Monitor
BABAR
Experiment running at Stanford Linear Accelerator studying the difference between matter and antimatter
Data selection:
Copy files from SLAC to Manchester (~5TB)
Select ~200 different streams using different criteria
Ship files of separate streams back to SLAC
SLAC computers overloaded with other processing tasks
Anticipate direct financial return if successful (common fund rebate to STFC)
Financial benefits (past)
BaBarGrid R011454 PPARC £138K
Testbed R011857 PPARC £10K
Grid Security
R011409 PPARC £461K
EGEE R013652 EU £112K
Tier2 operations
R011411 PPARC £311K
Financial benefits (future)
GridPP2+ PPARC ~£200K
GridPP3 PPARC £1.8M
EGEE3 EU £112K
Grants applied for
Expect approval though not at full amount requested
Summary
Particle Physics uses eScience
eScience benefits from Particle Physics
Manchester is leading in ideas and computer power thanks to bright people and strong support
Benefits in international recognition and research income
Long may it continue!