SC12 Home > SC12 Schedule > SC12 Presentation - Exploiting Network Parallelism for Improving Data Transfer Performance

SCHEDULE: NOV 10-16, 2012

When viewing the Technical Program schedule, on the far righthand side is a column labeled "PLANNER." Use this planner to build your own schedule. Once you select an event and want to add it to your personal schedule, just click on the calendar icon of your choice (outlook calendar, ical calendar or google calendar) and that event will be stored there. As you select events in this manner, you will have your own schedule to guide you through the week.

Exploiting Network Parallelism for Improving Data Transfer Performance

SESSION: SCinet Research Sandbox Presentations

EVENT TYPE: SCinet Research Sandboxes

TIME: 9:10AM - 9:30AM

SESSION CHAIR: Brian L. Tierney

Presenter(s):Dan Gunter, Raj Kettimuthu, Ezra Kissel, Martin Swany, Jun Yi, Jason Zurawski

ROOM:155-F2

ABSTRACT:
Enormous progress has been made in networking to support scientific use cases. Research and Education (R&E) backbones, including the Energy Sciences Network and Internet2, provide general-purpose services to allocate dedicated bandwidth. Many applications including bulk data transfer can often achieve significantly higher performance from virtually loss-free dedicated links than from opportunistic use. Bandwidth reservation introduces scheduling issues due to the relatively coarse grain of the allocations. Thus, links may appear full when in reality there may be multiple independent paths with aggregate available bandwidth to satisfy the request. Thus, applications may improve performance by using a combination of traditional "best effort" services and dynamic dedicated bandwidth with, e.g., OSCARS or Software Defined Networking (SDN). To meet this challenge, we propose using multiple paths for the same application transfer session. This may also facilitate performance with applications using multiple 10g NICs over 100g paths. GridFTP is widely used for bulk data movement. GridFTP transfers usually create parallel TCP streams, often along the same network topology. Intelligent session layer overlays, such as the Phoebus/XSP infrastructure, enable access to WAN acceleration techniques and transparent use of dynamic networking technologies. Embedded NetLogger monitoring can detect whether disks or networks are the limiting segment. The SCinet Research Sandbox will allow us to evaluate strategies that transparently map traffic to different paths, using available SDN capabilities. We will monitor the performance using our monitoring architecture called Periscope, and dynamically adapt the use of multipath capabilities based on knowledge of the end-to-end bottleneck.

Chair/Presenter Details:

Brian L. Tierney (Chair) - Lawrence Berkeley National Laboratory

Dan Gunter - Lawrence Berkeley National Laboratory

Raj Kettimuthu - Argonne National Laboratory

Ezra Kissel - University of Delaware

Martin Swany - Indiana University

Jun Yi - Argonne National Laboratory

Jason Zurawski - Internet2

Add to iCal  Click here to download .ics calendar file

Add to Outlook  Click here to download .vcs calendar file

Add to Google Calendarss  Click here to add event to your Google Calendar

Exploiting Network Parallelism for Improving Data Transfer Performance

SESSION: SCinet Research Sandbox Presentations

EVENT TYPE:

TIME: 9:10AM - 9:30AM

SESSION CHAIR: Brian L. Tierney

Presenter(s):Dan Gunter, Raj Kettimuthu, Ezra Kissel, Martin Swany, Jun Yi, Jason Zurawski

ROOM:155-F2

ABSTRACT:
Enormous progress has been made in networking to support scientific use cases. Research and Education (R&E) backbones, including the Energy Sciences Network and Internet2, provide general-purpose services to allocate dedicated bandwidth. Many applications including bulk data transfer can often achieve significantly higher performance from virtually loss-free dedicated links than from opportunistic use. Bandwidth reservation introduces scheduling issues due to the relatively coarse grain of the allocations. Thus, links may appear full when in reality there may be multiple independent paths with aggregate available bandwidth to satisfy the request. Thus, applications may improve performance by using a combination of traditional "best effort" services and dynamic dedicated bandwidth with, e.g., OSCARS or Software Defined Networking (SDN). To meet this challenge, we propose using multiple paths for the same application transfer session. This may also facilitate performance with applications using multiple 10g NICs over 100g paths. GridFTP is widely used for bulk data movement. GridFTP transfers usually create parallel TCP streams, often along the same network topology. Intelligent session layer overlays, such as the Phoebus/XSP infrastructure, enable access to WAN acceleration techniques and transparent use of dynamic networking technologies. Embedded NetLogger monitoring can detect whether disks or networks are the limiting segment. The SCinet Research Sandbox will allow us to evaluate strategies that transparently map traffic to different paths, using available SDN capabilities. We will monitor the performance using our monitoring architecture called Periscope, and dynamically adapt the use of multipath capabilities based on knowledge of the end-to-end bottleneck.

Chair/Presenter Details:

Brian L. Tierney (Chair) - Lawrence Berkeley National Laboratory

Dan Gunter - Lawrence Berkeley National Laboratory

Raj Kettimuthu - Argonne National Laboratory

Ezra Kissel - University of Delaware

Martin Swany - Indiana University

Jun Yi - Argonne National Laboratory

Jason Zurawski - Internet2

Add to iCal  Click here to download .ics calendar file

Add to Outlook  Click here to download .vcs calendar file

Add to Google Calendarss  Click here to add event to your Google Calendar