BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Denver
X-LIC-LOCATION:America/Denver
BEGIN:DAYLIGHT
TZOFFSETFROM:-0700
TZOFFSETTO:-0600
TZNAME:MDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0600
TZOFFSETTO:-0700
TZNAME:MST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20200129T163601Z
LOCATION:601
DTSTART;TZID=America/Denver:20191118T114000
DTEND;TZID=America/Denver:20191118T120500
UID:submissions.supercomputing.org_SC19_sess121_ws_pdsw103@linklings.com
SUMMARY:Active Learning-Based Automatic Tuning and Prediction of Parallel 
 I/O Performance
DESCRIPTION:Workshop\n\nActive Learning-Based Automatic Tuning and Predict
 ion of Parallel I/O Performance\n\nAgarwal, Singhvi, Malakar, Byna\n\nPara
 llel I/O is an indispensable part of scientific applications.  The current
  stack of parallel I/O contains many tunable parameters.  While changing t
 hese parameters can increase I/O performance many-fold, the application de
 velopers usually resort to default values because tuning is a cumbersome p
 rocess and requires expertise.  We propose two auto-tuning models, based o
 n active learning that recommend a good set of parameter values (currently
  tested with Lustre parameters and MPI-IO hints) for an application on a g
 iven system.  These models use Bayesian optimization to find the values of
  parameters by minimizing an objective function.  The first model runs the
  application to determine these values, whereas, the second model uses an 
 I/O prediction model for the same.  Thus the training time is significantl
 y reduced in comparison to the first model (e.g., from 800 seconds to 18 s
 econds).  Also both the models provide flexibility to focus on improvement
  of either read or write performance.  To keep the tuning process generic,
  we have focused on both read and write performance.  We have validated ou
 r models using an I/O benchmark (IOR) and 3 scientific application I/O ker
 nels (S3D-IO, BT-IO and GenericIO) on two supercomputers (HPC2010 and Cori
 ).  Using the two models, we achieve an increase in I/O bandwidth of up to
  11x over the default parameters.  We got up to 3x improvements for 37 TB 
 writes, corresponding to 1 billion particles in GenericIO.  We also achiev
 ed up to 3.2x higher bandwidth for 4.8 TB of non-contiguous I/O in BT-IO b
 enchmark.\n\nTag: Workshop Reg Pass, Big Data, Data Analytics, Data Manage
 ment, Storage\n\nRegistration Category: Workshop Reg Pass, Big Data, Data 
 Analytics, Data Management, Storage
URL:https://sc19.supercomputing.org/presentation/?id=ws_pdsw103&sess=sess1
 21
END:VEVENT
END:VCALENDAR