General issues for alignment

1
General issues for alignment

• Adlène Hicheur
• ATLAS Alignment meeting
• CERN, 25/05/05

2
Plan

• Processing chain for alignment
• Estimation of technical possibilities
• Various issues that need to be treated

3
Processing chain

• Alignment code assumed to run in the general
  chain during data processing, need to think about
  the consequences, e.g.
• Need to satisfy general C deontology and
  ATLAS requirements
• Need to have reasonable timing
• For the global ?2method, memory challenge
• Swap space for allocation of big matrix/vector
• Needs defined strategy
• Disk space for storage of big/matrix
• Should not be a problem, dedicated disk spaces,
  castor storage, etc
• For the robust method
• Need to think about the insertion of the
  iterative solving in the final scheme

4
Technical possibilities for space allocation

• Space allocation (aligning the whole geometry)
• Storage of symmetric big matrix vector 5GB
• If solving transition matrix during
  diagonalization 10GB
• Current machines do not allow more than 1-2GB
  (not to be used totally!)
• With 64bit architecture machine, would be Ok
• With current machines
• Fill different sub-matrices in parallel jobs
• If NxN matrix, fill p2 N/pxN/p sub-matrices
• e.g. for whole barrelendcap, N36000. if p6.
  Fill 36 6000x6000 sub-matrices which size is
  150MB.
• Just have to adapt the read method to gather all
  the bits before solving via parallel computing
  (shared memory for big matrix)

5
Other issues

• Event/track quality selector for alignment
• Benefit from scheduled studies on different data
  samples
• Muons coming from B?µX, W?µ?, Z?µµ,
• Minimum bias events (soft inelastic pp)
  dominated by charged pions. Select primary pions.
• Electrons bremsstrahlung?
• Common tool highly encouraged
• Survey information initial conditions
• Analysis of QA measurements
• Put numbers in exploitable format
• Include them in the alignment fits
• SCT Situation reasonable for endcap, but
  worrying for barrel need someone ASAP to
  analyze and check survey data (mounting robot),
  B3 is already at CERN!
• Read/Write/Update constants in DB
• Machinery to be validated

6
Questions on optimization of resources and
software

• Is the solving routine really the most optimized
  one?
• What is the compromise between disk place
  allocation and controlled solving?
• Parallel computing may consider improvements
  while learning about tools, e.g. is MPI the best
  choice as a message-passing library?
• How about testing ACS (developed by Syracuse
  Univ.)? See Park S-Y. Hariri S., Cluster
  Computing, 1999, vol. 2, no. 3, pp. 229-246(18)
  for performances comparison.

7
Conclusions

• Tour of different challenges opportunity to give
  a global view of ID alignment
• Where we stand alignment codes studies in Athena
  framework ? good basis of progress
• What we want