BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Access Bit Scanning for Page Replacement Policies
DTSTART:20260622T100000
DTEND:20260622T120000
DTSTAMP:20260523T201415Z
UID:5e20cb4164b13cfd44970942b365dd1422fe58b059a2106c0f02d147
CATEGORIES:Conferences - Seminars
DESCRIPTION:\nArunkrishna Annai Madalam Sivasubramanian  \nEDIC candidac
 y exam\nExam president: Prof. Thomas Bourgeat\nThesis advisor: Prof. Babak
  Falsafi\nCo-examiner: Prof. Sanidhya Kashyap\n\nAbstract\nModern workload
 s with memory footprints ranging from hundreds of gigabytes to several ter
 abytes have become widespread over the last decade. With memory accounting
  for nearly 50% of server costs in modern data centres\, provisioning larg
 e DRAM capacities significantly increases total cost of ownership (TCO). T
 iered memory architectures combine memories with different capacities\, la
 tencies\, and cost characteristics to improve cost efficiency while satisf
 ying application performance requirements. Application performance in thes
 e architectures is highly sensitive to page placement decisions. Consequen
 tly\, making informed placement decisions requires precise tracking of dat
 a access patterns and accurate identification of hot pages.\n\nExisting op
 erating systems commonly rely on hardware-maintained access bits in page t
 able entries to estimate memory access behaviour and identify hot data reg
 ions. They periodically scan the application's entire virtual address spac
 e to construct workload access profiles. However\, linear scanning of page
  table access bits does not scale for modern applications with terabyte-sc
 ale memory footprints. Since scanning overhead scales with the memory foot
 print\, a single full scan can take several seconds for workloads with lar
 ge memory footprints. Reducing scanning frequency lowers monitoring overhe
 ad but also decreases profiling accuracy and responsiveness to changing ac
 cess patterns\, resulting in a fundamental tradeoff between profiling accu
 racy and monitoring overhead. Therefore\, we need scalable\, low-overhead 
 mechanisms for access monitoring in next-generation memory management syst
 ems.\n\nIn this talk\, I will discuss recent approaches to scalable memory
  access profiling in large-memory systems. Specifically\, I will present w
 orks that use techniques such as adaptive sampling\, region-based monitori
 ng\, and hierarchical profiling of page table trees to reduce and bound sc
 anning overhead. \n\nSelected papers\ncoming soon
LOCATION:BC 010 https://plan.epfl.ch/?room==BC%20010
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR