skip to main content

DOE PAGESDOE PAGES

Title: Adaptive track scheduling to optimize concurrency and vectorization in GeantV

The GeantV project is focused on the R&D of new particle transport techniques to maximize parallelism on multiple levels, profiting from the use of both SIMD instructions and co-processors for the CPU-intensive calculations specific to this type of applications. In our approach, vectors of tracks belonging to multiple events and matching different locality criteria must be gathered and dispatched to algorithms having vector signatures. While the transport propagates tracks and changes their individual states, data locality becomes harder to maintain. The scheduling policy has to be changed to maintain efficient vectors while keeping an optimal level of concurrency. The model has complex dynamics requiring tuning the thresholds to switch between the normal regime and special modes, i.e. prioritizing events to allow flushing memory, adding new events in the transport pipeline to boost locality, dynamically adjusting the particle vector size or switching between vector to single track mode when vectorization causes only overhead. Lastly, this work requires a comprehensive study for optimizing these parameters to make the behaviour of the scheduler self-adapting, presenting here its initial results.
Authors:
 [1] ;  [2] ;  [3] ;  [1] ;  [4] ;  [1] ;  [5] ;  [6] ;  [4] ;  [1] ;  [4] ;  [4] ;  [1] ;  [7] ;  [8] ;  [1]
  1. European Organization for Nuclear Research (CERN), Geneva (Switzerland)
  2. Univ. of Catania and INAF, Catania (Italy)
  3. Univ. of Athens (Greece)
  4. Fermi National Accelerator Lab. (FNAL), Batavia, IL (United States)
  5. Univ. of Copenhagen (Denmark)
  6. Intel Corp., Santa Clara, CA (United States)
  7. Bhabha Atomic Research Center (BARC), Mumbai (India)
  8. National Technical Univ. of Ukraine, "Kyiv" Politechnic Institute (Ukraine)
Publication Date:
Report Number(s):
FERMILAB-CONF-14-590-CD
Journal ID: ISSN 1742-6588; 1372955
Grant/Contract Number:
AC02-07CH11359
Type:
Accepted Manuscript
Journal Name:
Journal of Physics. Conference Series
Additional Journal Information:
Journal Volume: 608; Journal Issue: 1; Journal ID: ISSN 1742-6588
Publisher:
IOP Publishing
Research Org:
Fermi National Accelerator Lab. (FNAL), Batavia, IL (United States)
Sponsoring Org:
USDOE Office of Science (SC), High Energy Physics (HEP) (SC-25)
Country of Publication:
United States
Language:
English
Subject:
72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS
OSTI Identifier:
1332185

Apostolakis, J., Bandieramonte, M., Bitzes, G., Brun, R., Canal, P., Carminati, F., Licht, J. C. De Fine, Duhem, L., Elvira, V. D., Gheata, A., Jun, S. Y., Lima, G., Novak, M., Sehgal, R., Shadura, O., and Wenzel, S.. Adaptive track scheduling to optimize concurrency and vectorization in GeantV. United States: N. p., Web. doi:10.1088/1742-6596/608/1/012003.
Apostolakis, J., Bandieramonte, M., Bitzes, G., Brun, R., Canal, P., Carminati, F., Licht, J. C. De Fine, Duhem, L., Elvira, V. D., Gheata, A., Jun, S. Y., Lima, G., Novak, M., Sehgal, R., Shadura, O., & Wenzel, S.. Adaptive track scheduling to optimize concurrency and vectorization in GeantV. United States. doi:10.1088/1742-6596/608/1/012003.
Apostolakis, J., Bandieramonte, M., Bitzes, G., Brun, R., Canal, P., Carminati, F., Licht, J. C. De Fine, Duhem, L., Elvira, V. D., Gheata, A., Jun, S. Y., Lima, G., Novak, M., Sehgal, R., Shadura, O., and Wenzel, S.. 2015. "Adaptive track scheduling to optimize concurrency and vectorization in GeantV". United States. doi:10.1088/1742-6596/608/1/012003. https://www.osti.gov/servlets/purl/1332185.
@article{osti_1332185,
title = {Adaptive track scheduling to optimize concurrency and vectorization in GeantV},
author = {Apostolakis, J. and Bandieramonte, M. and Bitzes, G. and Brun, R. and Canal, P. and Carminati, F. and Licht, J. C. De Fine and Duhem, L. and Elvira, V. D. and Gheata, A. and Jun, S. Y. and Lima, G. and Novak, M. and Sehgal, R. and Shadura, O. and Wenzel, S.},
abstractNote = {The GeantV project is focused on the R&D of new particle transport techniques to maximize parallelism on multiple levels, profiting from the use of both SIMD instructions and co-processors for the CPU-intensive calculations specific to this type of applications. In our approach, vectors of tracks belonging to multiple events and matching different locality criteria must be gathered and dispatched to algorithms having vector signatures. While the transport propagates tracks and changes their individual states, data locality becomes harder to maintain. The scheduling policy has to be changed to maintain efficient vectors while keeping an optimal level of concurrency. The model has complex dynamics requiring tuning the thresholds to switch between the normal regime and special modes, i.e. prioritizing events to allow flushing memory, adding new events in the transport pipeline to boost locality, dynamically adjusting the particle vector size or switching between vector to single track mode when vectorization causes only overhead. Lastly, this work requires a comprehensive study for optimizing these parameters to make the behaviour of the scheduler self-adapting, presenting here its initial results.},
doi = {10.1088/1742-6596/608/1/012003},
journal = {Journal of Physics. Conference Series},
number = 1,
volume = 608,
place = {United States},
year = {2015},
month = {5}
}