BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Improving the Scalability of HPC Applications by Separating Comput
 ation from Communication
DTSTART:20260721T160000Z
DTEND:20260721T170000Z
DTSTAMP:20260715T182000Z
UID:indico-event-915@events.cels.anl.gov
CONTACT:ALCFevents@alcf.anl.gov
DESCRIPTION:Register to join our July 21\, 2026\, webinar\, “Improving t
 he Scalability of HPC Applications by Separating Computation from Communic
 ation\,” to learn how a two-level MPI parallelization approach using one
 -sided communication can reduce communication bottlenecks\, simplify MPI p
 rogramming\, and enable extreme-scale application performance.\nFast progr
 ess in computer hardware poses a significant challenge for application dev
 elopers\, as hardware parallelism is increasing much faster than applicati
 ons can be adapted to use it effectively. When parallelism does not natura
 lly exist in an application\, it must be created. As applications scale\, 
 they often encounter communication bottlenecks caused by data dependencies
  among parallel processes.\nSeparation of computation from communication i
 s a distributed-memory parallelization approach designed to address this p
 roblem. This technique uses two-level MPI parallelization\, with one level
  performing intermediate data computation and the other handling data cons
 umption. The model maps naturally to one-sided MPI communication\, signifi
 cantly reducing programming complexity.\nComputation begins with all ranks
  in the first level computing their portion of the intermediate data. Once
  that work is complete\, each rank that consumes the data identifies which
  rank holds the required data and retrieves it through a one-sided MPI_Ge
 t. This lightweight communication layer resolves data dependencies between
  ranks and completes the MPI communication phase\, allowing the final data
 -consumption phase to proceed without interruption for data retrieval.\nBy
  redundantly computing intermediate data across multiple subcommunicators 
 and limiting MPI traffic to local subcommunicators\, this approach transfo
 rms the quadratically scaling all-to-all communication pattern into a line
 arly scaling block-diagonal communication matrix. Applied to the Fast Mult
 ipole Method\, which is known to be communication-bound\, this technique r
 educes communication cost to less than 1% of total time to solution at ful
 l machine scale on Aurora\, helping prepare the application for zettascale
  systems.\n \nVictor Anisimov is a Computational Scientist at the Argonn
 e Leadership Computing Facility. He holds a Ph.D. in Physical Chemistry fr
 om the Institute of Chemical Physics\, Russian Academy of Sciences (1997)\
 , followed by five years of computational chemistry software development a
 t Fujitsu\, where his team developed the linear-scaling semi-empirical qua
 ntum chemistry code LocalSCF. He conducted postdoctoral research at the Un
 iversity of Maryland\, Baltimore (2003–2008)\, and the University of Tex
 as at Houston (2008–2011)\, where he improved molecular dynamics methods
  and contributed to the CHARMM code.\nFrom 2011 to 2019\, at the National 
 Center for Supercomputing Applications at the University of Illinois Urban
 a-Champaign\, Dr. Anisimov supported petascale resource allocation teams o
 n the Blue Waters supercomputer\, optimized a variety of application codes
 \, and improved the performance of the coupled-cluster singles and doubles
  code in NWChem by a factor of two. He is the co-author\, with Dr. James J
 . P. Stewart\, of the textbook Introduction to the Fast Multipole Method.
  Dr. Anisimov specializes in performance optimization of molecular modelin
 g application codes.\n\nhttps://events.cels.anl.gov/event/915/
IMAGE;VALUE=URI:https://events.cels.anl.gov/event/915/logo-2834996557.png
LOCATION:Online
URL:https://events.cels.anl.gov/event/915/
END:VEVENT
END:VCALENDAR