Upload
aaron-salisbury
View
217
Download
0
Tags:
Embed Size (px)
Citation preview
Tony Doyle - University of Glasgow
UKUK
title.open ( ); revolution {execute};title.open ( ); revolution {execute};
LHC Computing ChallengeLHC Computing Challenge
Methodology?Methodology?
HHierarchical ierarchical IInformation in a nformation in a GGlobal lobal GGrid rid SSupernetupernet
Aspiration?Aspiration?
HIGGSHIGGS
DataGRID-UKDataGRID-UK
Aspiration?Aspiration?
ALLALL Data Intensive Computation Data Intensive Computation Teamwork
Tony Doyle - University of Glasgow
UKUK
OutlineOutline
Starting PointStarting Point The LHC The LHC
Computing Computing ChallengeChallenge
Data HierarchyData Hierarchy DataGRIDDataGRID Analysis Analysis
ArchitecturesArchitectures
GRID Data GRID Data ManagementManagement
Industrial Industrial PartnershipPartnership
Regional CentresRegional Centres Today’s WorldToday’s World Tomorrow’s WorldTomorrow’s World SummarySummary
Tony Doyle - University of Glasgow
UKUK
Starting PointStarting Point
Tony Doyle - University of Glasgow
UKUK
Starting PointStarting Point
“Current technology would not be able to scale data to such an extent, which is where the teams at Glasgow and Edinburgh Universities come in.The funding awarded will enable the scientists to prototype a Scottish Computing Centre which could develop the computing technology and infrastructure needed to cope with the high levels of data produced in Geneva, allowing the data to be processed, transported, stored and mined. Once scaled down, the data will be distributed for analysis by thousands of scientists around the world. The project will involve participation from Glasgow University's Physics & Astronomy and Computing Science departments, Edinburgh University's Physics & Astronomy department and the Edinburgh Parallel Computing Centre, and is funded by the Scottish Higher Education Funding Council's (SHEFC Joint Research Equipment Initiative). It is hoped that the computing technology developed during the project will have wider applications in the future, with possible uses in astronomy, computing science and genomics observation, as well as providing generic technology and software for the next generation Internet.”
Tony Doyle - University of Glasgow
UKUK
The LHC Computing ChallengeThe LHC Computing Challenge
Detector for ALICE experiment
Detector forLHCb experiment
Tony Doyle - University of Glasgow
UKUK
A Physics EventA Physics Event
Gated electronics response from a proton-proton collisionGated electronics response from a proton-proton collision Raw data: hit addresses, digitally converted charges and timesRaw data: hit addresses, digitally converted charges and times Marked by a unique code:Marked by a unique code:
Proton bunch crossing number, RF bucket Event number
Collected, Processed, Analyzed, Archived….Collected, Processed, Analyzed, Archived…. Variety of data objects become associated Event “migrates” through analysis chain:
may be reprocessed; selected for various analyses; replicated to various locations.
Tony Doyle - University of Glasgow
UKUK
LHC Computing ModelLHC Computing Model
Hierarchical, distributed tiersHierarchical, distributed tiers
GRID ties distributed resources together GRID ties distributed resources together
Tier-2
Tier-1
Tier-0 Dedicated or QoS Network Links
ScotGRID
CERN
CERN
Universities
RALRAL
Tony Doyle - University of Glasgow
UKUK
coordination required at collaboration and group levels
Data StructureData Structure
Raw DataRaw Data
Reconstruction
Data Acquisition
Level 3 trigger
Trigger TagsTrigger Tags
Event Summary Data
ESD
Event Summary Data
ESD Event Tags Event Tags
Physics Models
Monte Carlo Truth DataMonte Carlo Truth Data
MC Raw DataMC Raw Data
Reconstruction
MC Event Summary DataMC Event Summary Data MC Event Tags MC Event Tags
Detector Simulation
Calibration DataCalibration Data
Run ConditionsRun Conditions
Trigger System
Tony Doyle - University of Glasgow
UKUK
Physics AnalysisPhysics Analysis
ESD: Data or Monte CarloESD: Data or Monte Carlo
Event Tags Event TagsEvent Selection
Analysis Object DataAnalysis Object DataAnalysis Object DataAnalysis Object DataAnalysis Object Data
AOD
Analysis Object Data
AOD
Calibration DataCalibration Data
Analysis, Skims
Raw DataRaw Data
Tier 0,1Collaboration
wide
Tier 2Analysis
Groups
Tier 3, 4Physicists
Physics Analysis
Physics
Objects Physics
Objects
Physics
Objects
INC
RE
AS
ING
DA
TA
FLO
W
Tony Doyle - University of Glasgow
UKUK
ATLAS ParametersATLAS Parameters
Running conditions at startup:Running conditions at startup:
Raw event size ~2 MB (recently revised upwards...)Raw event size ~2 MB (recently revised upwards...)
2.7x102.7x1099 event sample event sample 5.4 PB/year, before data processing 5.4 PB/year, before data processing
““Reconstructed” events, Monte Carlo data Reconstructed” events, Monte Carlo data ~9 PB/year (2PB disk) ~9 PB/year (2PB disk)
CPU: ~2M SpecInt95 CPU: ~2M SpecInt95
CERN alone can handle only 1/3 of these resources
2005 2006 2007Average Luminosity (10^33) 0.1 1 10Trigger Rate (Hz) 100 270 400Physics Rate (Hz) 100 155 240Running (Equiv. Days) 14 100 100Physics Events (10^9) 0.1 2.7 2.4
Tony Doyle - University of Glasgow
UKUK
Data HierarchyData Hierarchy
““RAW, ESD, AOD, TAG”RAW, ESD, AOD, TAG”
RAWRAW Recorded by DAQRecorded by DAQTriggered eventsTriggered events
Detector digitiDetector digitissationation~2 MB/event~2 MB/event
ESDESDPseudo-physical information:Pseudo-physical information:
Clusters, track candidates Clusters, track candidates (electrons, muons), etc.(electrons, muons), etc.
Reconstructed Reconstructed informationinformation
~100 kB/event~100 kB/event
AODAOD
Physical informationPhysical information::Transverse momentum, Transverse momentum,
Association of particles, jets, Association of particles, jets, (best) id of particles,(best) id of particles,
Physical info for relevant “objects”Physical info for relevant “objects”
Selected Selected informationinformation
~10 kB/event~10 kB/event
TAGTAGAnalysis Analysis
informationinformation~1 kB/event~1 kB/eventRelevant information Relevant information
for fast event selectionfor fast event selection
Tony Doyle - University of Glasgow
UKUK
Testbed DataBaseTestbed DataBase
Object Model:Object Model:Atlas Simulated Raw Events Atlas Simulated Raw Events
bPEvent
bPEventObjVector
bPEventObj
bPSiDetector
bPSiDigit
bPMDT_Detector
bPMDT_Digit
bPCaloRegion
bPCaloDigit
bPTruthVertex
bPTruthTrack
System DB Raw Data DB1Raw Data DB2
...
Event Container Raw Data Container
PEvent #1 PEventObjeVector PEventObjVector :PEvent #2 PEventObjVector PEventObjVector :
PSiDetector PSiDigit ...PTRT_Detector PTRTDigit ...PMDT_Detector PMDT_Digit ...PCaloRigion PCaloDigit ...PTruthVertex PTruthTrack ... :
Tony Doyle - University of Glasgow
UKUK
LHC Computing ChallengeLHC Computing Challenge
Tier2 Centre ~1 TIPS
Online System
Offline Farm~20 TIPS
CERN Computer Centre >20 TIPS
RAL Regional Centre
US Regional Centre
French Regional Centre
Italian Regional Centre
InstituteInstituteInstituteInstitute ~0.25TIPS
Workstations
~100 MBytes/sec
~100 MBytes/sec
100 - 1000 Mbits/sec
•One bunch crossing per 25 ns
•100 triggers per second
•Each event is ~1 Mbyte
Physicists work on analysis “channels”
Each institute has ~10 physicists working on one or more channels
Data for these channels should be cached by the institute server
Physics data cache
~PBytes/sec
~ Gbits/sec or Air Freight
Tier2 Centre ~1 TIPS
Tier2 Centre ~1 TIPS
~Gbits/sec
Tier Tier 00
Tier Tier 11
Tier Tier 33
Tier Tier 44
1 TIPS = 25,000 SpecInt95
PC (1999) = ~15 SpecInt95
ScotGRID++ ~1 TIPS
Tier Tier 22
Tony Doyle - University of Glasgow
UKUK
e.g. MySQL database daemon
Basic 'crash-me' and associated tests
Access times for basic insert, modify, delete, update database operations e.g.
(on 256Mbyte, 800MHz Red Hat 6.2 linux box)
Database Access BenchmarkDatabase Access Benchmark
350k data insert operations 149 seconds
10k query operations 97 seconds
350k data insert operations 149 seconds
10k query operations 97 seconds
Many applications require database functionalityMany applications require database functionality
Currently favoured HEP DataBase applicatione.g. BaBar, ZEUS software
Tony Doyle - University of Glasgow
UKUK
CPU Intensive ApplicationsCPU Intensive Applications
Numerically intensive simulations:Numerically intensive simulations: Minimal input and output data
ATLAS Monte Carlo (gg H bb)228 sec/3.5 Mb event on 800 MHz linux
box
Standalone physics applications:
1. Simulation of neutron/photon/electron interactions for 3D detector design2. NLO QCD physics simulation
Compiler Speed (MFlops)Fortran (g77) 27C (gcc) 43Java (jdk) 41
Compiler Tests:
Tony Doyle - University of Glasgow
UKUK
Network Monitoring PrototypeNetwork Monitoring Prototype
Tools:Java
Analysis Studio
overTCP/IP
InstantaneousCPU Usage
ScalableArchitecture
Individual Node Info.
Tony Doyle - University of Glasgow
UKUK
Analysis ArchitectureAnalysis Architecture
Converter
Algorithm
Event DataService
PersistencyService
DataFiles
AlgorithmAlgorithm
Transient Event Store
Detec. DataService
PersistencyService
DataFiles
Transient Detector
Store
MessageService
JobOptionsService
Particle Prop.Service
OtherServices
HistogramService
PersistencyService
DataFiles
TransientHistogram
Store
ApplicationManager
ConverterConverter
The Gaudi Framework - developed by LHCb
- adopted by ATLAS (Athena)
Tony Doyle - University of Glasgow
UKUK
GRID ServicesGRID Services
Grid ServicesGrid Services Resource Discovery Scheduling Security Monitoring Data Access Policy
Athena/Gaudi ServicesAthena/Gaudi Services Application manager
“Job Options” service
Event persistency service
Detector persistency
Histogram service
User interfaces
Visualization
DatabaseDatabase Event model
Object federations
Extensible interfaces and
protocols being specified
and developed:
Tools: 1. UML
2. Java
Protocols: 1. XML
2. MySQL DataGRID Toolkit
3. LDAP}
Tony Doyle - University of Glasgow
UKUK
Virtual Data ScenarioVirtual Data Scenario
Example analysis scenario:Example analysis scenario: Physicist issues a query from Athena for a Monte Carlo dataset
Issues: How expressive is this query? What is the nature of the query: declarative Creating new queries and language
Algorithms are already available in local shared libraries
An Athena service consults an ATLAS Virtual Data Catalog
Consider possibilities:Consider possibilities: TAG file exists on local machine (e.g. Glasgow)
Analyze it
ESD file exists in a remote store (e.g. Edinburgh) Access relevant event files, then analyze that
RAW File no longer exists (e.g. RAL) Regenerate, re-reconstruct, re-analyze !!! GRID Data
Management
Tony Doyle - University of Glasgow
UKUK
GlobusGlobus
Tony Doyle - University of Glasgow
UKUK
GlobusGlobus
DataGRIDToolKit
Tony Doyle - University of Glasgow
UKUK
GRID Data ManagementGRID Data Management
Goal: develop middle-ware infrastructure to manage petabyte-scale data
Replica Manager
Data Mover
Data Accessor
Storage Manager
Castor HPSS
Data Locator
Meta Data Manager
Local Filesystem
Query Optimisation &Access Pattern Manag.
Secure Region
High Level Services
Medium Level Services
Core ServicesService levels reasonably well defined
Identify Key AreasWithin Software
Structure
Tony Doyle - University of Glasgow
UKUK
5 areas for development5 areas for development Data Accessor - hides specific storage system requirements.
Mass Storage Management group. Replication - improves access by wide-area caching. Globus
toolkit offers sockets and a communication library, Nexus. Meta Data Management - data catalogues, monitoring
information (e.g. access pattern), grid configuration information, policies. MySQL over Lightweight Directory Access Protocol (LDAP) being investigated.
Security - ensuring consistent levels of security for data and meta data.
Query optimisation - “cost” minimisation based on response time and throughput Monitoring Services group.
Identifiable UKContributions
RAL
Identifying Key AreasIdentifying Key Areas
RAL
Tony Doyle - University of Glasgow
UKUK
AstroGridAstroGrid
WP1 PROJECT MANAGEMENT
WP2 REQUIREMENTS ANALYSIS : existing functionality and future requirements; community consultation
WP3 SYSTEM ARCHITECTURES: benchmark and implement
WP4 GRID-ENABLE CURRENT PACKAGES : implement and test performance
WP5 DATABASE SYSTEMS : requirements analysis and implementation; scalable federation tools.
WP6 DATA MINING ALGORITHMS : requirements analysis, development and implementation
WP7 BROWSER APPLICATIONS : requirements analysis and software development
WP8 VISUALISATION : concepts and requirements analysis, software development.
WP9 INFORMATION DISCOVERY : concepts and requirements analysis, software development
WP10 FEDERATION OF KEY CURRENT DATASETS : e.g.. SuperCOSMOS, INT-WFS, 2MASS, FIRST, 2dF
WP11 FEDERATION OF NEXT GENERATION OPTICAL-IR DATASETS : esp. Sloan, WFCAM
WP12 FEDERATION of HIGH ENERGY ASTROPHYSICS DATASETS : esp. Chandra, XMM
WP13 FEDERATION of SPACE PLASMA and SOLAR DATASETS : esp. SOHO, Cluster, IMAGE
WP14 COLLABORATIVE DEVELOPMENT OF VISTA, VST, and TERAPIX PIPELINES
WP15 COLLABORATION PROGRAMME WITH INTERNATIONAL PARTNERS
WP16 COLLABORATION PROGRAMME WITH OTHER DISCIPLINES
Emphasis on High LevelGUIs etc
WP 1 Grid Workload Management A.Martin-QMW (0.5)
WP 2 Grid Data Management A.Doyle-Glasgow (1.5)
WP 3 Grid Monitoring services R.Middleton-RAL (1.8)
WP 4 Fabric Management A.Sansum-RAL (0.5)
WP 5 Mass Storage Management J.Gordon-RAL (1.5)
WP 6 Integration Testbed D.Newbold-Bristol (3.0)
WP 7 Network Services P.Clarke-PPNCG/UCL (2.0)
WP 8 HEP Applications N/A (?) (4.0)
WP 9 EO Science Applications ( c/o R.Middleton-RAL ) (0.0)
WP 10 Biology Applications ( c/o P.Jeffreys-RAL ) (0.1)
WP 11 Dissemination P.Jeffreys-RAL (0.1)
WP 12 Project Management R.Middleton-RAL (0.5)
ReplicationFragmentation
Emphasis on Low LevelServices etc
Tony Doyle - University of Glasgow
UKUK
Testbed = Learning by ExampleTestbed = Learning by Example
+Cloning
SRIF Expansion
= expansion of open source ideas
“GRID Culture”
Tony Doyle - University of Glasgow
UKUK
missionmission to accelerate the exploitation of simulation by to accelerate the exploitation of simulation by industry, commerce and academia industry, commerce and academia
45 staff, £2.5M turnover - externally funded45 staff, £2.5M turnover - externally funded solve business problems - not sell technologysolve business problems - not sell technology
PartnershipImportant
Tony Doyle - University of Glasgow
UKUK
Industrial PartnershipIndustrial Partnership
pingping
service
ping
monitor
WAN
LAN
Adoption of OPENIndustry Standards
+OO Methods
Industry ResearchCouncil Inspiration:
Data-IntensiveComputation
Tony Doyle - University of Glasgow
UKUK
Regional CentresRegional Centres
SRIF Infrastructure
Grid Data Management
SecurityMonitoring
Networking
Local Perspective:Consolidate
Research Computing
Optimisation of Number of Nodes?4-5?
Relative size dependent on funding dynamics
Global Perspective:V. Basic Grid Skeleton
Regional Expertise Model?
Tony Doyle - University of Glasgow
UKUK
Today’s WorldToday’s World
Istituto Trentino Di Cultura
Helsinki Institute of Physics
Science Research Council
SARA
Tony Doyle - University of Glasgow
UKUK
Tomorrow’s WorldTomorrow’s World
CR2
AC12
AC13
AC14
Istituto Trentino Di
CulturaHelsinki Institute of
PhysicsScience Research Council
AC7
AC8
AC9
AC10
AC11
CR3
AC15
AC16
AC17
CR4
SARA
AC18
AC19
CR5
AC20
AC21
CR6
CO
Tony Doyle - University of Glasgow
UKUK
SummarySummary
General Engagement (£=OK)General Engagement (£=OK) Mutual Interest (Mutual Interest (ScotGRIDScotGRID
Example)Example) Emphasis on Emphasis on
DataGrid Core Development (e.g. Grid Data Management)
“CERN” lead + Unique UK Identity Extension of Open Source Idea “Grid
Culture” = Academia + Industry Multidisciplinary Approach =
University + Regional Basis Use of Existing Structures (e.g. EPCC,
RAL) Hardware Infrastructure via SRIF +
Industrial Sponsorship Now LHC
Grid Data Management
SecurityMonitoring
Networking
Detector for ALICE experiment
Detector forLHCb experiment
ScotG
RI
D