BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//wp-events-plugin.com//7.4.12//EN
TZID:Asia/Kolkata
X-WR-TIMEZONE:Asia/Kolkata
BEGIN:VEVENT
UID:233@cds.iisc.ac.in
DTSTART;TZID=Asia/Kolkata:20261119T110000
DTEND;TZID=Asia/Kolkata:20261119T120000
DTSTAMP:20261105T080506Z
URL:https://cds.iisc.ac.in/events/ph-d-thesis-defense-leveraging-replicati
 on-for-integration-of-fault-tolerance-and-high-performance-in-mpi-applicat
 ions/
SUMMARY:Ph.D. Thesis Defense: Leveraging Replication for Integration of Fau
 lt Tolerance and High Performance in MPI Applications
DESCRIPTION:DEPARTMENT OF COMPUTATIONAL AND DATA SCIENCES\nPh.D. Thesis Def
 ense\n\n\n\nSpeaker: Mr. Sarthak Joshi\nS.R. Number: 06-18-01-10-12-21-1-1
 9304\nTitle: Leveraging Replication for Integration of Fault Tolerance and
  High Performance in MPI Applications\nResearch Supervisor: Prof. Sathish 
 Vadhiyar\nDate &amp\; Time : November 19\, 2026 (Thursday)\, 11:00 A.M.\nV
 enue : #102\, CDS Seminar Hall\n\n\n\nABSTRACT\nFaults in high-performance
  systems are expected to be very frequent in the current exascale computin
 g era. In this thesis\, we have defined FTHP-MPI (Fault Tolerant and High 
 Performance MPI)\, a fault-tolerant MPI library that augments checkpoint/r
 estart with replication to provide resilience from failures. Our library i
 s designed to provide fault tolerance in a native MPI library that does no
 t provide support for fault tolerance. This enables users to transparently
  and seamlessly implement fault tolerance into an MPI application while st
 ill utilizing the underlying high performance communication implementation
 s of the native MPI library. We have implemented efficient parallel commun
 ication techniques that involve replicas. Our framework deals with the uni
 que challenges of integrating support for checkpointing and replication in
 cluding handling the differences in the number of processes at the point o
 f checkpoint creation and restart and implementing transparent checkpointi
 ng and replication while handling differences and overlaps in virtual memo
 ry addresses across the processes. We have extensively tested our library 
 under both failure-free and failure conditions using real executions with 
 three benchmark applications\, namely\, HPCG\, CloverLeaf mini-application
  and the Particle-In-Cell Simulation\, scaling up to 16K processes. We hav
 e observed that under failure-free conditions\, our library only incurs a 
 3-8% overhead due to the presence of replicas when the number of computati
 onal processes is the same. We simulated failure conditions by killing pro
 cesses either randomly at intervals based on a Weibull Distribution or usi
 ng super computing logs from Tsubame-3 supercomputing system. We show that
  our library can outperform pure checkpointing by 13-19% with the same num
 ber of processes across both cases\, even when half the processes do redun
 dant work.\n\nWe have also developed mechanisms in our framework to provid
 e adaptive partial replication to mitigate the efficiency loss due to empl
 oying additional resources for replication. Our framework utilizes failure
  prediction to protect a large number of processes using only a small numb
 er of replicas. However\, imperfect failure predictors can result in appli
 cation interruptions\, necessitating frequent checkpointing. In a first su
 ch work\, we have developed a formulation for optimal checkpointing interv
 al\, taking into account the recall metric of the failure predictor and th
 e replication degree\, to minimize the checkpointing overheads. We have te
 sted adaptive replication using a variety of simulated failure predictors 
 with different precision and recall values. We observe that adaptive repli
 cation outperforms checkpointing and full replication by 56.7% and 27.6%\,
  respectively even with relatively bad failure predictors with 50% precisi
 on and recall. We also developed a simulation framework to project the imp
 act of replication-based fault tolerance for modern exascale systems.\n\nW
 e show that\, at the MTBF values observed in modern exascale systems\, whi
 le replication without any failure predictions is not a viable approach\, 
 adaptive replication can still outperform pure checkpointing even with a l
 ow recall when running large-scale executions. Furthermore\, replication-b
 ased approaches become more suitable for fault tolerance when compared to 
 pure checkpointing as MTBF values are expected to decrease further\, and r
 eductions in checkpoint overheads by continuous advancements in checkpoint
 ing can also be leveraged by adaptive replication implementations.\n\nFina
 lly\, we also implemented replication-based fault tolerance framework in M
 PI-IO\, the first of its kind. Our approach utilizes the replicas as addit
 ional resources that can share the work and improve the efficiency while s
 till providing fault tolerance. We observe that in a setup with a signific
 ant share of I/O\, our fault-tolerant library with replication can reduce 
 the execution time by 44-47%.\n\n\n\nALL ARE WELCOME
CATEGORIES:Events,Thesis Defense
END:VEVENT
BEGIN:VTIMEZONE
TZID:Asia/Kolkata
X-LIC-LOCATION:Asia/Kolkata
BEGIN:STANDARD
DTSTART:20251119T110000
TZOFFSETFROM:+0530
TZOFFSETTO:+0530
TZNAME:IST
END:STANDARD
END:VTIMEZONE
END:VCALENDAR