BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20191206T142456Z
LOCATION:Substanz 1\, 2
DTSTART;TZID=Europe/Stockholm:20180627T113000
DTEND;TZID=Europe/Stockholm:20180627T120000
UID:isc_hpc_ISC High Performance 2018_sess151_pap145@linklings.com
SUMMARY:Machine Learning Based Parallel I/O Predictive Modeling: A Case St
 udy on Lustre File Systems
DESCRIPTION:Research Paper\n\nMachine Learning Based Parallel I/O Predicti
 ve Modeling: A Case Study on Lustre File Systems\n\nMadireddy, Balaprakash
 , Carns, Robert, Ross...\n\nParallel I/O hardware and software infrastruct
 ure is a key contributor to performance variability for applications runni
 ng on large-scale HPC systems. This variability confounds efforts to predi
 ct application performance for characterization, modeling, optimization, a
 nd job scheduling.  We propose a modeling approach that improves predictiv
 e ability by explicitly treating the variability and by leveraging the sen
 sitivity of application parameters on performance to group applications wi
 th similar characteristics. \nWe develop a Gaussian Process-based machine 
 learning algorithm to model I/O performance and its variability as a funct
 ion of application and file system characteristics. We demonstrate the eff
 ectiveness of the proposed approach using data collected from the Edison s
 ystem at the National Energy Research Scientific Computing Center (NERSC).
  The results show that the proposed sensitivity-based models are better at
  prediction when compared to application-partitioned or unpartitioned mode
 ls. We highlight modeling techniques that are robust to the outliers that 
 can occur in production parallel file systems. Using the developed metrics
  and modeling approach, we provide insights into the file system metrics t
 hat have a significant impact on I/O performance.\n\nPasses: Conference Pa
 ss, AI/Machine Learning/Deep Learning, File Systems, Performance Analysis 
 and Optimization\n\nTag: Conference Pass, AI/Machine Learning/Deep Learnin
 g, File Systems, Performance Analysis and Optimization
URL:https://2018.isc-program.com/presentation/?id=pap145&sess=sess151
END:VEVENT
END:VCALENDAR