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:20200129T163556Z
LOCATION:704-706
DTSTART;TZID=America/Denver:20191118T143500
DTEND;TZID=America/Denver:20191118T150500
UID:submissions.supercomputing.org_SC19_sess136_ws_canopie107@linklings.co
 m
SUMMARY:Enabling HPC workloads on Cloud Infrastructure Using Kubernetes Co
 ntainer Orchestration Mechanisms
DESCRIPTION:Workshop\n\nEnabling HPC workloads on Cloud Infrastructure Usi
 ng Kubernetes Container Orchestration Mechanisms\n\nBeltre, Saha, Govindar
 aju, Younge, Grant\n\nContainers offer a broad array of benefits, includin
 g a consistent lightweight runtime environment through OS-level virtualiza
 tion, as well as low overhead to maintain and scale applications with high
  efficiency. Moreover, containers are known to package and deploy applicat
 ions consistently across varying infrastructures. Container orchestrators 
 manage a large number of containers for microservices based cloud applicat
 ions.  However, the use of such service orchestration frameworks toward HP
 C workloads remains relatively unexplored.\n\nIn this paper, we study the 
 potential use of Kubernetes on HPC infrastructure for use by the scientifi
 c community.  We directly compare both its features and performance agains
 t Docker Swarm and bare metal execution of HPC applications. Herein, we de
 tail the configurations required for Kubernetes to operate with containeri
 zed MPI applications, specifically accounting for operations such as (1) u
 nderlying device access, (2)  inter-container communication across differe
 nt hosts, and (3) configuration limitations.\n\nThis evaluation quantifies
  the performance difference between representative MPI workloads running b
 oth on bare metal and containerized orchestration frameworks with Kubernet
 es, operating over both Ethernet and InfiniBand interconnects. Our results
  show that Kubernetes and Docker Swarm can achieve near bare metal perform
 ance over RDMA communication when high performance transports are enabled.
  Our results also show that Kubernetes presents overheads for several HPC 
 applications over TCP/IP protocol. However, Docker Swarm's throughput is n
 ear bare metal performance for the same applications.\n\nTag: Workshop Reg
  Pass, Containers, HPC\n\nRegistration Category: Workshop Reg Pass, Contai
 ners, HPC
URL:https://sc19.supercomputing.org/presentation/?id=ws_canopie107&sess=se
 ss136
END:VEVENT
END:VCALENDAR