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:20260522T150117Z
LOCATION:D170
DTSTART;TZID=America/Chicago:20181111T153000
DTEND;TZID=America/Chicago:20181111T155200
UID:submissions.supercomputing.org_SC18_sess149_ws_mchpc112@linklings.com
SUMMARY:On the Applicability of PEBS-Based Online Memory Access Tracking f
 or Heterogeneous Memory Management at Scale
DESCRIPTION:Aleix Roca Nonell (Barcelona Supercomputing Center, Polytechni
 c University of Catalonia); Balazs Gerofi (RIKEN); Leonardo Bautista-Gomez
  (Barcelona Supercomputing Center); Dominique Martinet (Atomic Energy and 
 Alternative Energies Commission (CEA)); Vicenç Beltran Querol (Barcelona S
 upercomputing Center); and Yutaka Ishikawa (RIKEN)\n\nOperating systems ha
 ve historically had to manage only a single type of memory device. The imm
 inent availability of heterogeneous memory devices based on emerging memor
 y technologies confronts the classic single memory model and opens a new s
 pectrum of possibilities for memory management. Transparent data movement 
 based on access patterns of applications is a desired feature to hide the 
 complexity of memory management to end users. However, capturing memory ac
 cess patterns of an application at runtime comes at a cost, which is parti
 cularly challenging for large scale parallel applications that may be sens
 itive to system noise.In this work, we focus on the access pattern profili
 ng. We study the feasibility of using Intel’s Processor Event Based Sampli
 ng (PEBS) feature to record memory accesses by sampling at runtime and stu
 dy the overhead at scale. We have implemented a custom PEBS driver in the 
 IHK/McKernel lightweight multi-kernel operating system, one of whose advan
 tages is minimal system interference due to the lightweight kernel’s simpl
 e design compared to other OS kernels such as Linux. We present the PEBS o
 verhead of a set of scientific applications and show the access patterns i
 dentified in noise sensitive HPC applications. Our results show that clear
  access patterns can be captured with 10% overhead in the worst case when 
 running on up to 128k CPU cores (2,048 Intel Xeon Phi Knights Landing node
 s). We conclude that online memory access profiling using PEBS at large sc
 ale is promising for memory management in heterogeneous memory environment
 s.\n\nTag: Memory, NVRAM, Parallel Programming Languages, Libraries, and M
 odels\n\nRegistration Category: Workshop Reg Pass\n\nSession Chairs: Ron B
 rightwell (Sandia National Laboratories); Maya Gokhale (Lawrence Livermore
  National Laboratory (LLNL)); Xian-He Sun (Illinois Institute of Technolog
 y); and Yonghong Yan (University of North Carolina, Charlotte)\n\n
END:VEVENT
END:VCALENDAR