BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/New_York
X-LIC-LOCATION:America/New_York
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20210402T160556Z
LOCATION:Track 5
DTSTART;TZID=America/New_York:20201112T113200
DTEND;TZID=America/New_York:20201112T115500
UID:submissions.supercomputing.org_SC20_sess211_ws_pdsw108@linklings.com
SUMMARY:Toward On-Demand I/O Forwarding in HPC Platforms
DESCRIPTION:Workshop\n\nToward On-Demand I/O Forwarding in HPC Platforms\n
 \nBez, Zanon Boito, Miranda, Nou, Cortes...\n\nI/O forwarding is an establ
 ished and widely-adopted technique in HPC to reduce contention and improve
  I/O performance in the access to shared storage infrastructure. On such m
 achines, this layer is often physically deployed on dedicated nodes, and t
 heir connection to the clients is static. Furthermore, the increasingly he
 terogeneous workloads entering HPC installations stress the I/O stack, req
 uiring tuning and reconfiguration based on the applications' characteristi
 cs. Nonetheless, it is not always feasible in a production system to explo
 re the potential benefits of this layer under different configurations wit
 hout impacting clients. In this paper, we investigate the effects of I/O f
 orwarding on performance by considering the application's I/O access patte
 rns and system characteristics. We aim to explore when forwarding is the b
 est choice for an application, how many I/O nodes it would benefit from, a
 nd whether not using forwarding at all might be the correct decision. To g
 ather performance metrics, explore, and understand the impact of forwardin
 g I/O requests of different access patterns, we implemented FORGE, a light
 weight I/O forwarding layer in user-space. Using FORGE, we evaluated the o
 ptimal forwarding configurations for several access patterns on MareNostru
 m 4 (Spain) and Santos Dumont (Brazil) supercomputers. Our results demonst
 rate that shifting the focus from a static system-wide deployment to an on
 -demand reconfigurable I/O forwarding layer dictated by application demand
 s can improve I/O performance on future machines.\n\nTag: Big Data, Data A
 nalytics, Compression, and Management, Data Movement, File Systems and I/O
 , Storage\n\nRegistration Category: Workshop Reg Pass
END:VEVENT
END:VCALENDAR