BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Computer Laboratory Computer Architecture Group Me
 eting
SUMMARY:Using Processor Hardware Counters in Picking the O
 ptimal Work-Stealing Policy - Milos Puzovic\, Univ
 ersity of Cambridge
DTSTART;TZID=Europe/London:20110624T160000
DTEND;TZID=Europe/London:20110624T170000
UID:TALK31954AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/31954
DESCRIPTION:Randomised work-stealing is a distributed dynamic 
 load balancing scheme\nthat has been proven to be 
 optimal with tight space bounds for a large\nnumbe
 r of recursive task-parallel programs. Under work-
 stealing when a\nparent thread spawns a child\, wo
 rker will make parent thread available\nto idle wo
 rkers to steal it. However\, if parallelism manife
 sted in\nprogram is iterative then the steals from
  idle workers will be serialised\nand therefore ha
 d a negative impact on scalability of the\nschedul
 er. In this case we would have been much better of
 f if we chose\nto continue executing the parent th
 read and make a child thread available\nfor steali
 ng.\n\nIn this talk I will introduce tools I have 
 developed to characterise different\ntypes of task
 -parallel benchmarks and present the insights how 
 to\nexploit now ubiquitous hardware counters to dy
 namically pick and switch\nbetween the optimal pol
 icies during the program's lifetime.\n
LOCATION:FW26\, Computer Laboratory
CONTACT:Prof Simon Moore
END:VEVENT
END:VCALENDAR