BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Chicago
X-LIC-LOCATION:America/Chicago
BEGIN:DAYLIGHT
TZOFFSETFROM:-0600
TZOFFSETTO:-0500
TZNAME:CDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0500
TZOFFSETTO:-0600
TZNAME:CST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260522T150121Z
LOCATION:D168
DTSTART;TZID=America/Chicago:20181112T121500
DTEND;TZID=America/Chicago:20181112T122000
UID:submissions.supercomputing.org_SC18_sess140_ws_isav101@linklings.com
SUMMARY:Scheduling for In-machine Analytics: Data Size Is Important
DESCRIPTION:Valentin Honore (University of Bordeaux, French Institute for 
 Research in Computer Science and Automation (INRIA)) and Guillaume Aupy an
 d Brice Goglin (French Institute for Research in Computer Science and Auto
 mation (INRIA), University of Bordeaux)\n\nWith the goal of performing exa
 scale computing, the importance of I/O management becomes increasingly cri
 tical to maintain system performance.  While the computing capacities of m
 achines are getting higher, the I/O capabilities of systems do not follow 
 the same trend.  To address this issue, the HPC community proposed new sol
 utions such as online in-machine analysis to overcome the limitations of b
 asic post-mortem data analysis where the data have to be stored on the Par
 allel File System (PFS) first to be processed later.\n\nIn this paper, we 
 propose to study different scheduling strategies for in-machine analytics.
  Our goal is to extract the most important features of analytics that dire
 ctly determine the efficiency  of scheduling strategies. To do so, we prop
 ose a memory-constraint modelization for in-machine\nanalysis. It automati
 cally determines hardware resource partitioning and proposes scheduling po
 licies for simulation and analysis.  We evaluate our model through simulat
 ions and observe that it is critical to base scheduling decisions on the m
 emory needs of each analytics. We also note unexpected behaviors from whic
 h we deduce that modeling the in-machine paradigm for HPC applications req
 uires deep understanding of task placement, data movement and hardware par
 titioning.\n\nTag: Data Analytics, Data Management, Visualization\n\nRegis
 tration Category: Workshop Reg Pass\n\nSession Chairs: Earl P.N. Duque (In
 telligent Light); Nicola Ferrier (Argonne National Laboratory (ANL), Unive
 rsity of Chicago); Kenneth Moreland (Oak Ridge National Laboratory (ORNL))
 ; and Matthew Wolf (Oak Ridge National Laboratory (ORNL))\n\n
END:VEVENT
END:VCALENDAR