Summary of Computing Section of Technical Proposal

1
Summary of Computing Section of Technical Proposal

• Data Flow Model
• Computing Requirements
• Computing Infrastructure
• Software Strategy
• Project Organisation and Management
• Manpower estimates and costs

2
Status of Documents

• Draft of Computing Section is available - 5 pages
• Based on four Computing Notes containing more
  details
• LHC-B Computing Tasks Requirements
• LHC-B Computing Model
• LHC- B Software Strategy
• LHC-B Project Plan for Computing
• Drafts of notes available on Web - still being
  revised

3
(No Transcript)
4
Data Flow Model

• Algorithms used for Level 2/3 triggers similar to
  those employed in full reconstruction. Issues are
  
• speed, reliability
• calibration and alignment in real-time
• output of L2 and L3 used by full reconstruction
• Size of data store and access speeds 2-3 orders
  of magnitude higher than current experiments and
  similar to other LHC experiments
• Raw data written to storage at 20 MB/s.
• Similar amount of reconstruction information (14
  MB/s)
• In total capability of storing gt 0.4 PB of
  data/year at 40 MB/s
• Transparent access to data store by nearly all
  tasks

5
Computing Requirements

• Estimates of CPU requirements, input/output data
  volumes based on
• simulation - program exists so good estimates
• Assumptions on evolution of algorithms
• optimisation (e.g. shower parameterisation)
• increasing complexity (more detail)
• new frameworks like GEANT4 (30 improvement)
• reconstruction - partial information on pattern
  recognition
• Extrapolations from existing experiments (need
  input from HERA-B)
• analysis algorithms - less well known but needs
  are smaller
• Some numbers are targets as opposed to
  benchmarks
• for example, goals for L2/L3 are 10/200 msec (on
  1000 Mips CPU)

6
Dedicated Installed Processing Power
7
Data Storage Requirements
8
Computing Infrastructure

• Issues are
• Strategy for evolution of computing model
• Timescales for investment in computing resources
• Scalability of cpu farms needed for cpu intensive
  processing
• Handling of Petabytes of data stored in a central
  database
• Equal access to data for all collaboration
  institutes

9
Evolution of Computing Infrastructure

• Steady investment in desktop systems
• Preparation Phase (1998 - 2000)
• 1998 - need is 1000 Mips and 2 TB of data
• Use public facilities both inside and outside
  CERN
• Increase of 50/year in requirements for
  simulation and analysis
• Impact of test-beam?
• Implementation Phase (2001 - 2003)
• significant increase in our needs (TDRs, full MC,
  testbeam)
• invest in private (SHIFT-like) facilities (end
  2000/beginning of 2001)
• Commissioning Phase (2004 - 2005)
• assembly of full-scale facilities
• financing scheme

10
Data Storage Model

• All event data stored in a single Object Database
• Storage/retrieval managed by a hierarchical mass
  storage system
• Assume 10 of data stored on disk
• Study options for access of data from any
  institute
• CERNtric model - all data stored at and accessed
  from CERN
• Regional centres - data distributed between CERN
  and home labs
• Cache (part of) data at each institute
• Depends on technology (network), tariffs,
  logistics, politics

11
Software Strategy

• Objectives
• quality in software ( trigger, prompt
  reconstruction.)
• performance - trigger latencies, CPU for bulk
  processing
• improve on
• knowledge of PEOPLE involved
• the organisation of the development PROCESS
• the TECHNOLOGY used
• Approach
• use appropriate engineering practices
• stress importance of architecture - adherence to
  standards
• build high quality components (manpower
  intensive)
• re-use components wherever possible (manpower
  efficient)
• use commercial products when appropriate
• participate in common (LHC-wide) projects
• plan well - encourage all members of
  collaboration to participate

12
Software StrategyTechnology

• Specialised tools that help building software for
  all life-cycle activities
• project management (MSProject, communication
  (web), workflow)
• verification (inspection, testing) - Purify,
  Logiscope
• configuration management (code and documents of
  all sorts)
• Technology for life-cycle phases
• TP states our intention is to adopt Object
  Technologies
• OOA (analysis), OOD (design), ODBMS (database),
  C/Java (language) integration standards
  (OMG/CORBA, ActiveX/DCOM, RMI/Javabeans)
• large investment by software industry -
  commercial tools and products widely available
  (GUIs, distributed systems)
• widespread adoption within HEP
• GEANT4 - new simulation framework re-engineered
  using OO
• Event Store/Objectivity
• Replacement of CERNLIB - OpenGL, IrisExplorer
  (analysis framework)
• Adoption by other experiments (BaBar, STAR,
  ATLAS/CMS,ALICE..)

13
Benefits of OO

• OO evolved out of addressing issues of
  programming-in-the-large
• Objects are basis for reusable modules
• Communication by message passing helps to define
  interfaces between modules and external systems
• Design essential features of an object that
  distinguish it from all other objects - defines
  crisp boundaries (Abstraction)
• All internal implementation details are hidden -
  manage complexity (Encapsulation)
• Reuse of well designed/tested modules (objects)
  gives better quality and leads to high
  productivity
• Partitioning of work into domains is much easier

14
Drawbacks of OO

• Field is still developing rapidly and some
  technologies/products may be superceded
• Culture change is necessary and , in general,
  people hate this
• Significant costs associated with training and
  re-education
• OO may not be the last word in software
  engineering

15
Migration Policy

• Steps are as follows
• Build up a suitable programming environment
  (e.g. C, UML, Rose)
• Develop frameworks for simulation, reconstruction
  and analysis
• impetus will come mid 98 with release of GEANT4
  and LHC toolkits
• Embark on intensive training programme
• Minimise legacy software - hence set an
  aggressive schedule
• Manpower is an important issue
• consolidation of SICB development
• need extra (skilled) effort

16
Steering Group

• Composition - coordinator plus one rep from each
  project
• Tasks - Coordination, Planning, Resources

Computing Facilities
Recon- struction
Analysis
Simulation
DAQ
Controls
OPS
Software Eng.Group

• Farms
• Desktop
• Storage
• Network
• Operating
• System

• Level 2 FW
• Level 3 FW
• Recon FW
• Calibration
• Production

• Framewk
• Tools

• GEANT4
• Framewk
• Tools
• Production

• Methods
• Tools
• Code
• Manag.
• Quality
• Document.
• Training
• Licenses
• Collab.
• Tools

• Event
• Builder
• Readout
• Network
• Interfaces
• Links
• Crates
• DAQware

• DCS
• LHC
• Safety
• Run
• Control

• Operations
• Consoles
• Shift Crew
• Enviroment

Re-usable Components

• Data Management Event Store, Geometry, Database
  Utilities, ODBMS
• Architecture Frameworks, Component model,
  Distributed system
• Toolkits GUI, Histograms, Communications
• Utilities data quality monitoring, event
  display, bookkeeping

17
Links to Sub-detector Groups
Application Project (e.g. Reconstruction)
RICH Computing Team

• Project Leader
• Vertex
• RICH
• Inner Tracker
• Outer Tracker
• ECAL
• HCAL
• MUON
• Trigger L0
• Trigger L1
• Trigger L2/L3

18
Life-Cycle Phases

• Preparation Phase ( now until end of 2000)
  Learning
• collect requirements and develop functional
  specifications of subsystems
• evaluate hardware technologies
• build prototypes
• Implementation Phase (start 01 until end
  03) Building
• make technology choices
• engineer sub-systems
• Commissioning Phase (start 04 until end 04)
  Testing
• install
• unit test, integration tests
• tests under realistic loads (bulk data, realistic
  real-time tests)
• Operation Phase (start 05 until physics goals
  archived) Running
• support
• adapt and improve

19
Manpower Estimates
Group
98 99 00 01 02 03 04 05
Comments
Steering Group
1 1
2 2 2 2 2 2
DAQ
4 6
6 10 10 10 10 8
1-2/
subdetector
Controls
1 1
1 2 3 3 3
2
Common Project
Operations
0 0
0 0 0 1 2 4
Simulation
3 3
3 3 3 3 2
2
1-2/sub-detector
Reconstruction
2 2
2 3 3 3 3
3
1-2/sub-detector
Analysis
2 2
2 3 3 4 4
4
interactive applications
Re-usable components
2 2
2 7 7 7 4 4
Software Engineering
2 2
2 5 5 5 4
4
Common Project
Computing Facilities
2 2
2 5 5 5 8
8
TOTALS
19 21 22 40 41 43
42 41
20
Cost Estimate
Initial Investment Cost
Item
Units
Unit Cost
1
Total Cost
6
CPU (
Mips
)
1x10
3 SFr
3.0
MSFr
2
Disk
(TB)
42
12
kSFr
0.5
MSFr
Tape (TB)
420
1
kSFr
0.5
MSFr
Total
4.0
MSFr
Notes
1. Taken from industry supplied extrapolations to
the year 2005
2. Assume 10 of total data taken will reside on
disk
Annual Investment Costs
Item
Units
Unit Cost
Total Cost
Desktop CPU
100
100 SFr/month
100
kSFr
Software
(LHC,OS)
100
kSFr
CPU
500
kSFR
Disk
200
kSFr
Tape
500
kSFr
Total
1400
kSFr
21
(No Transcript)