BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Pandia: comprehensive contention-sensitive thread placement - Dani
 el Goodman (Oracle)
DTSTART:20170330T140000Z
DTEND:20170330T143000Z
UID:TALK71651@talks.cam.ac.uk
CONTACT:Liang Wang
DESCRIPTION:In this talk we introduce Pandia is a system for modelling the
  performance of in-memory parallel workloads. Pandia generates a descripti
 on of a workload from a series of profiling runs\, and combines this with 
 a description of the machine's hardware to model the workload's performanc
 e over different thread counts and different placements of those threads.\
 n\nThe approach is “comprehensive” in that it accounts for contention 
 at multiple resources such as processor functional units and memory channe
 ls. The points of contention for a workload can shift between resources as
  the degree of parallelism and thread placement changes. Pandia accounts f
 or these changes and provides a close correspondence between predicted per
 formance and actual performance. Testing a set of 22 benchmarks on 2 socke
 t Intel machines fitted with chips ranging from Sandy Bridge to Haswell we
  see median differences of 1.05\\% to 0\\% between the fastest predicted p
 lacement and the fastest measured placement\, and median errors of 8\\% to
  4\\% across all placements.\n\nPandia can be used to optimize the perform
 ance of a given workload---for instance\, identifying whether or not multi
 ple processor sockets should be used\, and whether or not the workload ben
 efits from using multiple threads per core. In addition\, Pandia can be us
 ed to identify opportunities for reducing resource consumption where addit
 ional resources are not matched by additional performance---for instance\,
  limiting a workload to a small number of cores when its scaling is poor.\
 n\nBio: Daniel Goodman is a researcher at Oracle Labs Cambridge where he w
 orks on runtime systems. Prior to this he has worked as an RA at Mancheste
 r University and Oxford University looking at dataflow programming and abs
 tractions that make HPC computing more accessible.
LOCATION:FW26\, Computer Laboratory\, William Gates Building
END:VEVENT
END:VCALENDAR