BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Chicago
X-LIC-LOCATION:America/Chicago
BEGIN:DAYLIGHT
TZOFFSETFROM:-0600
TZOFFSETTO:-0500
TZNAME:CDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0500
TZOFFSETTO:-0600
TZNAME:CST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260522T150120Z
LOCATION:D163
DTSTART;TZID=America/Chicago:20181112T140000
DTEND;TZID=America/Chicago:20181112T142500
UID:submissions.supercomputing.org_SC18_sess142_ws_pdsw105@linklings.com
SUMMARY:Pufferbench: Evaluating and Optimizing Malleability of Distributed
  Storage
DESCRIPTION:Nathanael Cheriere (IRISA, ENS Rennes); Matthieu Dorier (Argon
 ne National Laboratory); and Gabriel Antoniu (French Institute for Researc
 h in Computer Science and Automation (INRIA))\n\nMalleability is the prope
 rty of an application to be dynamically rescaled at run time. It requires 
 the possibility to dynamically add or remove resources to the infrastructu
 re without interruption. Yet, many Big Data applications cannot benefit fr
 om their inherent malleability, since their colocated distributed storage 
 system is not malleable in practice. Commissioning or decommissioning stor
 age nodes is generally assumed to be slow, as such operations have typical
 ly been designed for maintenance only. New technologies, however, enable f
 aster data transfers. Still, evaluating the performance of rescaling opera
 tions on a given platform is a challenge in itself: no tool currently exis
 ts for this purpose.\n\nWe introduce Pufferbench, a benchmark for evaluati
 ng how fast one can scale up and down a distributed storage system on a gi
 ven infrastructure and, thereby, how viably can one implement storage mall
 eability on it. Besides, it can serve to quickly prototype and evaluate me
 chanisms for malleability in existing distributed storage systems. We vali
 date Pufferbench against theoretical lower bounds for commission and decom
 mission: it can achieve performance within 16% of them. We use Pufferbench
  to evaluate in practice these operations in HDFS: commission in HDFS coul
 d be accelerated by as much as 14 times! Our results show that: (1) the lo
 wer bounds for commission and decommission times we previously established
  are sound and can be approached in practice; (2) HDFS could handle these 
 operations much more efficiently; most importantly, (3) malleability in di
 stributed storage systems is viable and should be further leveraged for Bi
 g Data applications.\n\nTag: I/O, Storage\n\nRegistration Category: Worksh
 op Reg Pass\n\n
END:VEVENT
END:VCALENDAR