BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/New_York
X-LIC-LOCATION:America/New_York
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20210402T160558Z
LOCATION:Track 5
DTSTART;TZID=America/New_York:20201111T120000
DTEND;TZID=America/New_York:20201111T123000
UID:submissions.supercomputing.org_SC20_sess195_ws_mchpc109@linklings.com
SUMMARY:Performance Potential of Mixed Data Management Modes for Heterogen
 eous Memory Systems
DESCRIPTION:Workshop\n\nPerformance Potential of Mixed Data Management Mod
 es for Heterogeneous Memory Systems\n\nEffler, Jantz, Jones\n\nMany high-p
 erformance systems now include different types of memory devices within th
 e same compute platform to meet strict performance and cost constraints. S
 uch heterogeneous memory systems often include an upper-level tier with be
 tter performance, but limited capacity, and lower-level tiers with higher 
 capacity, but less bandwidth and longer latencies for reads and writes. To
  utilize the different memory layers efficiently, current systems rely on 
 hardware-directed, memory-side caching or they provide facilities in the o
 perating system (OS) that allow applications to make their own data-tier a
 ssignments. Since these data management options each come with their own s
 et of trade-offs, many systems also include mixed data management configur
 ations that allow applications to employ hardware- and software-directed m
 anagement simultaneously, but for different portions of their address spac
 e.\n\nDespite the opportunity to address limitations of stand-alone data m
 anagement options, such mixed management modes are under-utilized in pract
 ice, and have not been evaluated in prior studies of complex memory hardwa
 re. In this work, we develop custom program profiling, configurations, and
  policies to study the potential of mixed data management modes to  outper
 form hardware- or software-based management schemes alone. Our experiments
 , conducted on an Intel Knights Landing platform with high-bandwidth memor
 y, demonstrate that the mixed data management mode achieves the same or be
 tter performance than the best stand-alone option for five memory intensiv
 e benchmark applications (run separately and in isolation), resulting in a
 n average speedup compared to the best stand-alone policy of over 10%, on 
 average.\n\nRegistration Category: Workshop Reg Pass
END:VEVENT
END:VCALENDAR