Frank van Lingen fvlingencacr'caltech'edu

1

Current Future work

• Frank van Lingen (fvlingen_at_cacr.caltech.edu)
• California Institute of Technology

2
The MONARC 2 Simulation Framework

• The aim was to continue and improve the
  development of the MONARC simulation framework.
  
• The new framework was redesign based on the
  MONARC experience and it has been developed,
  improved and tested by
• Iosif Legrand (CALTECH)
• Ciprian Dobre (UPB)
• Corina Stratan (UPB)
• Ramiro Voicu (UPB/CALTECH)
• Frank van Lingen (CALTECH)
• Kathrin Paschen (CERN)
• Andrew Fritz (Houston University)
• http//monalisa.cacr.caltech.edu/MONARC/

3
Monarc extensions I developed

• Use of object collections within the simulation
  (e.g. give me objects of type X within the range
  of a,b and process them with application Y).
• Being able to specify a direct acyclic graph of
  processes that need to be performed in that order
• Introduction of failure of resources such as
  servers, and cpus for an amount of time.
• Introducing a job execution module that would
  execute jobs on particular resources based on an
  execution strategy (currently this functionality
  is empty)
• Replica location. Depending on the resource you
  would like to execute your job, you can select
  different replicas.
• Replica management. If data becomes "hot" you
  might need to replicate this data. What are the
  criteria for this?
• I use the term job as an application that reads
  data, process data, and write data.

4
Monarc extensions cont.

• Added methods that would allow you to add a
  "cost" to resources you want to use, and costs
  for the jobs you want to process. Currenlty
  "empty", but idea is to implement different cost
  functions. You can then attach a "budget" to
  different groups that submit these jobs. Using
  budgets and resource costs you can "prioritize"
  work of particular groups without completely
  blocking the work of others.

5
Grid Analysis Environment

• Being involved in discussions related to an
  architecture for interactive analysis on the grid
  This activity is lead by Michael Thomas from
  Caltech and started after the workshop held at
  Caltech 2 months ago
• see
• http//pcbunn.cacr.caltech.edu/GAE/workshop_agenda
  .html )

6
Short term work

• Currently I am trying do develop scenarios based
  on initial ideas for the data challenge within
  CMS (see last slide). The ideas on this slide are
  not final and might change during the next
  months. Furthermore, within the CMS data
  challenge there is a very fixed and predetermined
  way of processing data, and not as dynamic as
  sometimes has been envisioned within a grid.

7
Long term work

• Idea is to develop different schedulers, cost
  functions, replication strategies and job
  execution strategies and see what the effect is
  on a grid in combination with job patterns
  injected into that grid. This could eventually
  lead to a more "self organizing grid"
• looking at physics analysis where people from all
  over the world request resources for their jobs.
  This process will be more "chaotic" than the data
  challenge. Some constraints that are related to
  this are
• You do not want to have one person to consume all
  the resources (hence the introduction of budgets)
• You do not want to have people to wait
  indefinetely for their jobs to be finished.
  (their should be a degree of fairness)
• I would be interested to know if there other
  persons doing similar work on simulation (within
  or outside griphyn)