ESnet Abilene Cooperative Measurements

Introduction

There is a desire to show that network performance between ESnet sites and Internet2 sites served by Abilene is equivalent to network performance across one of the networks or the other. Network performance is complex to measure and involves many factors including latency, bandwidth, loss and jitter. The first step towards this goal will be to setup an infrastructure to measure latency between 3 ESnet sites and 3 Internet2/Abilene sites using OWAMP.

OWAMP Latency Measurements

Abilene has been using OWAMP for some time to measure latency across their backbone links and ESnet has used ping to measure latency across their backbone. The focus of this effort is to measure end to end performance between ESnet sites and Abilene participants.

Preliminary data from owping end-to-end tests are available. Note, the tools used to measure, archive, and display this data are under development. You can safely assume that the missing data on graphs at this site represents development of the data collection system, not faults in the underlying network.

This preliminary data collection is gathered by an ad-hoc system that runs OWPING -i 5 -c 20 between the various servers every 5 minutes.

LBL Measurements
LBL to NCSU	LBL to SDSC	LBL to OSU	LBL to FERMI	LBL to BNL
FERMI Measurements
FERMI to LBL	FERMI to BNL	FERMI to NCSU	FERMI to SDSC	FERMI to OSU
BNL Measurements
BNL to LBL	BNL to FERMI	BNL to NCSU	BNL to SDSC	BNL to OSU
NCSU Measurements
NCSU to FERMI	NCSU to LBL	NCSU to OSU	NCSU to SDSC
SDSC Measurements
SDSC to FERMI	SDSC to LBL	SDSC to BNL	SDSC to NCSU
Ohio State University Measurements
OSU to FERMI	OSU to LBL	OSU to BNL	OSU to NCSU
CERN Measurements
CERN to FERMI	CERN to LBL	CERN to BNL

Interesting Observations from Past Data.

• LBL to NCSU

  This path crosses ESnet, Abilene, NCREN & NCNI.

  One interesting issue in this data is that the latency increased by approximately 400 usec for a week in early May.
  

  Joe,
  Regarding the question about the 400 usec jump in latency on the LBL-NCSU path on Tuesday morning: we moved the circuit connecting my router in the lab to the NCNI edge router to a different path on our metro DWDM system at that time. The optical distance increased by about 60 km. I think it's pretty cool that the measurements were accurate enough to pick up the delta.

  BTW, we may move that circuit again in a few weeks, to dark fiber. If that happens the optical distance will be decreased by about 170 km.
  John M

• Owamp measurement reliance on NTP

  This data collection does illustrate one interesting detail about the measurement system. At approximately 5:15 PST Monday May 10th the OWAMP reported precision estimates derived from NTP dropped to significantly less than the apparent error. They stayed at the artifically low values for about 14 hours. We are working on understanding why this happens.

Collaborators

The results displayed on this page are the work of many people.
Most of them are on the i2-esnet@internet2.edu mailing list.

Some of the key people are:

• Jin Guojun @ LBL
• John Moore & Chintan Desai @ NCSU
• Kevin Walsh @ SDSC
• Jeff Boote, Eric Boyd & Guy Almes @ Internet2
• Chin Guok, Joe Metzger & the rest of the ESnet team.
• Phil Demar & Darryl Wohlt @ FNAL
• Ohio State University

This page is maintained by Joe Metzger.
Please send me any comments or updates.