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:20260522T150121Z
LOCATION:D163
DTSTART;TZID=America/Chicago:20181111T163000
DTEND;TZID=America/Chicago:20181111T170000
UID:submissions.supercomputing.org_SC18_sess159_ws_indis103@linklings.com
SUMMARY:Analysis of CPU Pinning and Storage Configuration in 100 Gbps Netw
 ork Data Transfer
DESCRIPTION:Se-young Yu, Jim Chen, Joe Mambretti, and Fei Yeh (Internation
 al Center for Advanced Internet Research (iCAIR), Northwestern University)
 \n\nA common bottleneck for high-speed network data transfers is lack of C
 PU resources. A number of techniques and solutions have been proposed to r
 educe CPU load for data transfer. One can optimize the core affinity setti
 ngs in their Non-Uniform Memory Access (NUMA) system and use NVMe over Fab
 rics to avoid CPU bottlenecks in high-speed network data transfers. Our as
 sumption is that binding processes to the local processor improves the ove
 rall performance of the high-speed network data transfers compared to bind
 ing the processes to actual cores or leaving them unbounded. Furthermore, 
 using NVMe over Fabrics reduces the CPU utilization more with a lower numb
 er of processors. To evaluate these assumptions, we performed a series of 
 experiments with different core affinity and storage settings. We found ev
 idence that binding processes to the local processor instead of the cores 
 improve the file transfer performance for most of the use-cases and NVMe o
 ver Fabrics is more efficient in transferring files compared to traditiona
 l file transfers in Local Area Networks (LANs). We were able to achieve th
 e maximum SSD performance threshold using 32 transfer processes with tradi
 tional file transfers while using 8 processes with NVMe over Fabrics and r
 educed CPU utilization.\n\nTag: Architectures, Networks, Security\n\nRegis
 tration Category: Workshop Reg Pass\n\nSession Chairs: Ilya Baldin (Thomas
  Jefferson National Accelerator Facility); Paola Grosso (University of Ams
 terdam, Netherlands); Mary Hester (Dutch National Institute for Subatomic 
 Physics); and Michelle Zhu (Montclair State University)\n\n
END:VEVENT
END:VCALENDAR