Pegasus:

1
Pegasus Particle-physics Engagement with the
Grid A Socio-technical Usability Study.
Dr. Will Venters (Principle Investigator) Informat
ion Systems and Innovation Group Department of
Management The London School of Economics and
Political Science w.venters_at_lse.ac.uk This
research was undertaken as part of Pegasus EPSRC
Grant No EP/D049954/1 www.pegasus.lse.ac.uk
2
The Pegasus Project

• To observe the way in which GridPP comes to
  develop, shape and exploit grid infrastructure.
• Focus on practice on the mundane.
• Advisory Group
• Prof. Tony Doyle, Prof. Elaine Ferneley, Prof.
  Steve Lloyd, Prof. Wanda Orlikowski,Dr Susan
  Scott.

Yingqin
Will
Mark Lancaster(UCL)
Tony
Avgousta
3
Thank you!

• Venters, W., T. Cornford, A. Kyriakidou, Y.
  Zheng. (2009) "Grids' Knowledge Infrastructure A
  case study of the LCG computing Grid at CERN
  through a practice lens." 5th International
  conference on e-Social Science, Cologne.
• Kyriakidou, A. W. Venters (2009) "Distributed
  large-scale systems development Exploring the
  collaborative development of the particle physics
  Grid." 5th International conference on e-Social
  Science, Cologne.
• Kyriakidou, A. Venters, W (2009) "Clusters Of
  Competence In The Distributed Development Of
  Grids A Particle Physics Communitys Response To
  The Difficulties Of Global Systems Development.,
  IFIP 9.5 Workshop, Athens, Greece.
• Kyriakidou, A. Venters, W (2008) "Distributed
  development of large-scale global systems
  Exploring the collaborative practices of particle
  physicists as they develop a Grid for the LHC",
  presented at IFIP 8.2 OASIS workshop 2008, Paris,
  France.
• Venters, W., Cornford, T., Zheng, Y (2008)
  "Grids A Knowledge Infrastructure requiring a
  Knowledge Infrastructure", presented at UK
  e-Science All Hands Meeting 2008, Edinburgh
• Zheng, Y., W. Venters and T. Cornford (2007)
  "Agility, Improvisation, or Enacted Emergence?"
  Proceedings of the 28th International Conference
  on Information Systems (ICIS 2007), Montreal,
  Canada. Best Paper Finalist.
• Venters,W., Zheng,Y., Cornford,T. (2007)
  Collaborative construction of grid technology
  Researching the usability of Grids within
  Particle Physics. Third International Conference
  on e-Social Science, Ann Arbor, Michigan US.
• Zheng, Y., W. Venters, T. Cornford (2007)
  "Distributed Development and Scaled Agility
  Improvising a Grid for Particle Physics", London
  School of Economics Working Paper Series - 163,
  London http//is2.lse.ac.uk/wp/pdf/wp163.pdf
  ISSN 1472-9601
• Kyriakidou, A. Venters, W (2007) The
  Multi-Disciplinary Development of Collaborative
  Grids The Social Shaping of a Grid for
  Healthcare. 15th European Conference on
  Information Systems, St Gallen, Switzerland.
  Nominated for Best Paper. Available Here.
• Scott,SV, Venters,W (2007) The Practice of
  e-Science and e-Social Science Method, Theory,
  and Matter. IFIP 8.2. Virtuality and
  Virtualization. Crowston, K., Sieber, S., Wynn,
  E. (Eds.), Vol. 236ISBN 978-0-387-73024-0,
  Springer.
• Venters,W., Cornford,T. (2006) Introducing
  Pegasus An ethnographic research project
  studying the use of Grid technologies by the UK
  particle Physics Community. Second International
  Conference on e-Social Science, 28-30 June 2006,
  Manchester, UK. Available here.
• Uribe, L.M. (2007) Socio-technical elements of
  e-Research and libraries. MSc Disseratation,
  Information Systems and Innovation Group, London
  School of Economics, supervised by Dr. W. Venters.

4
Pegasus future

• Framework of guidance on Grid development/deployme
  nt.
• Paradoxes scaled agility
• Knowledge Infrastructure of Grids.
• Activity-theory of distributed systems
  development practice.
• EPSRC bid to study GridPP through deployment and
  use

5
Overview of talk
1) Grandmother
2) Egg
Insert in mouth instruct to suck
I would value your feedback/critique email me
on address below!! w.venters_at_lse.ac.uk
6
GridPP3 - Deployment

• Deployment
• Very exciting time!
• Political and consequential
• Users and developers wrestle with the technology
  
• Technology becomes

7
What is ICT?

• Not designed to solve problems
• Both technology and problems are politically
  socially constructed!
• Relevant social groups
• Conflict then stabilization

Robert Moses
Winner, L. "Do artifacts have politics?," in The
Social Shaping of Technology, D. Mackenzie and J.
Wajcman (eds.), Open University Press,
Maidenhead, 1999, pp. 28-40.
8
Social Construction
Bijker, W. Of Bicycles, Bakerlites and Bulbs
Toward a theory of Socio-Technical Change MIT
Press, Cambridge, MA, 1995.
9
The Grid is

• A Physicists perspective
• just a processing machine that sits there
• CMS
• Powerful organisation in its own right.
• Members produce software to support members.
• Crab (CMS Remote Analysis Builder)
• to simplify the process of creation and
  submission of CMS analysis jobs into a Grid
• crab.cfg

10
Running a CRAB job

• Operating Crab (and LCG) within the CMS
  community
• People learnby getting a working script from
  someone and starting from there and trying to
  run it and trying to modify it to suit there
  needs.
• FAQs, documentation, Blogs and Wikis are used and
  developed.

11
Resistance by the technology

• Inevitably things go wrong
• Zero return codes Job-success was always Zero,
  no matter what happened - Impossible to
  monitor what is going wrong
• Crab gets blamed there is a school of thought
  that the last letter should be changed to P
• Unclear support (GGUS) No local systems
  administrators.
• Crab creates a barrier invariably problems
  fall between Grid and CMS. So Grid people dont
  know what I am talking about and CMS software
  people say it is a Grid problem.
• Physicists focus on the goal of physics
  analysisnot developing or improving the LCG grid.

12
Technology changes practice

• New work-practices technology
• Just kill problematic jobs which take too long
  to run
• Being a systems administrator for particle
  physics must be hell, they will want different
  systems for everything, and they are all quite
  savvy, they will tune everything.
• CMS community CRAB reflects these problemsand
  these ad-hoc solutions

13
CMS Improvising new practice

• Grids are coordinated resource sharing (Foster
  Kesselman 1998). Workload Manager to coordinate
  this.
• VOs can ban explicitly sites so there is a
  white-list and black-list. The Workload Manager
  is forced to consider this list.
• BL and WL are features for testing/management
• But CMS software aims to schedule jobs onto
  resources according to the policy and priorities
  of CMS (CMS Workbook my emphasis)
• Exploit middleware facility for CMS physicists

If you want/need to select/deselect some site,
you can use Ce_black_list (refuse access
to all the listed CEs, allow all
others) Ce_white_list (allow access only to
those CEs listed) FAQ details of CRAB.cfg
options (truncated)
14

• In practice the way I operate is I find out
  where my data is and tell this Crab tool where to
  go.
• Zero codes in that case I quite often try and
  send the job somewhere else, not use that
  particular Grid site
• the only way it works is if you find out where
  the data is and they you tell the thing to send
  it there.
• Doing this releases more time to make physics
  analysis happen
• Physicists understand that this causes problems
  elsewhere for Grid systems admin in particular.

15
Relevant Social Group SysAdmin

• Grid is shared with other LHC experiments and
  other sciences.
• The biggest problems today is managing such a
  big infrastructure where there are so many sites
  and not all the sites are managed with the same
  level or body of quality.
• Lack of data analysis for testing (lack of
  demand).
• Zero codes make sense cannot be expected to
  understand failures (Abbas 2004)
• CERN has a large amount of CEs and SEs available,
  well managed and with lots of the data locally.
• People feel they do not know who supports what

Abbas, A. Grid Computing A Practical Guide to
Technology and Applications Charles River Media,
MA, 2004.
16
SysAdmin

• Experiments develop independent monitoring
  systems () of the infrastructure as they see it.
  Not as the managers see it. And they develop
  client code that is able to do as much as
  possible in the automatic way of clean-up
• If there was a user sitting in isolation they
  would have to use what we provide. But they all
  work for experiments with lots of influence and
  resources and everything. And which occasionally,
  possibly often, have very high influence in some
  of the sites as well, they can ask the sites to
  install various services. So they can bypass
  stuff.
• But if this continues workload management might
  be scrapped, because it would be investment in
  something useless
• CMS is not alone in this - ATLAS and pilot jobs

17
So what? Why is this relevant?

• Just add another technological fix???
• Deployment of Information Infrastructure as
  socially constructed by relevant social groups
• Widely distributed systems
• No management control
• Users enrol powerful technology reflective of
  their practices
• Different technology providers
• Poorly understood politics

18
Lessons

• Co-aligning technical and social elements with
  the practice of relevant social groups (RSG).
• Consider RSG work practices and motivation in
  aspects of Grid design.
• Consider VOs as powerful agents of technological
  change.
• Design support and documentation to reflect the
  RSGs interpretation.
• Take seriously the role of sysadmin in
  socially-constructing infrastructure.

19
Electronic Prescription Service (EPS) The
Electronic Prescription Service will enable
prescribers - such as GPs and practice nurses -
to send prescriptions electronically to a
dispenser (such as a pharmacy) of the patients
choice.
Project CFH004 Evaluation of EPS.
20
(No Transcript)