BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Global Parallel Computation\, Networks and Geometry - Peter Braam 
 (Braam Research LLC\, USA)
DTSTART:20160301T160000Z
DTEND:20160301T170000Z
UID:TALK64557@talks.cam.ac.uk
CONTACT:Greg Willatt
DESCRIPTION:We discuss implementations of distributed algorithms for a bro
 ad family of linear algebra problems which lead to perfect load balancing 
 of data movement\, storage and computation.  The algorithms were likely fi
 rst discovered by Karmarkar in the 1990’s leveraging finite projective g
 eometries.  From this approach\, network topologies emerge which exhibit r
 esilience\, very low diameters and high cross sectional bandwidth.  A few 
 studies of these algorithms in software have shown significant speedup ove
 r current practice\, subject to proper locality of ingest data.  Silicon p
 hotonics may allow for effective implementation of the networks.  While it
  has received little attention\, this broad mixture of ideas appears to fo
 rm a viable system architecture for parallel computing\, but it also raise
 s questions about implementation methodology and performance studies of pa
 rallel algorithms.
LOCATION:Martin Ryle Seminar Room\, Kavli Institute
END:VEVENT
END:VCALENDAR