Traceroute Analysis of Transpacific Network

To: Yohkoh Colleagues

From: R. C. Canfield

Date: 22 January 1997

At MSU we recently did a study that tells a lot about the transpacific network. Using a simple unix shell script and cron, we gathered data from the unix traceroute command once an hour for about a month, from early December to early January 1997. We studied connections between MSU (sxt4.physics.montana.edu) and three destination hosts:

  • bigbear.caltech.edu at Big Bear Solar Observatory, California
  • isass0.solar.isas.ac.jp at ISAS, Sagamihara, Japan
  • sag.space.lockheed.com at Lockheed Martin ATC, California

    Traceroute sends out test packets and tracks them from gateway to gateway, determining the time for them to return (if they do return) along the routes taken. We found end to end (ETE) times for test packets sent from MSU, the fraction that return (reachability), and the detailed route information, and analyzed them superficially.

    It is useful to study our results for both ISAS and SAG. Along the most common routes, the average ETE time to ISAS is about 200 ms, whereas to SAG it is about 60 ms. The number of hops is similar. The reachability of ISAS averages only about 80 percent, and at some hours, only about 60 percent. In contrast, the reachability of SAG averages about 95 percent, and about half of the time (midnight to noon in Montana), it is 100 percent.

    Consider the full ISAS delay matrix for a typical route from MSU to ISAS, and compare it to the full SAG delay matrix for a typical route from MSU to SAG. Note that each row of the matrix corresponds to a gateway along the route (listed in Appendix B ). The explanation of Internet routers and lines between ISAS and the US from Miura san in (Appendix A) helps understand the geographic location of some of these gateways. From detailed identification of each row of the matrix with a gateway on the list in Appendix B, it becomes clear that the dramatic appearance of very long delays at all times during the day (red color in the ISAS delay matrix) coincides with the Japan - US 6 Mbps line (between #12 and #13 in Appendix B). The matrix also shows that another hop, which is in the US (between #10 and #11) is very flakey. Although the same flakey site exists between MSU and SAG as between MSU and ISAS, the overall situation is much better.

    If this traceroute stuff is familiar to you, then I'd appreciate some feedback on the following issue. Various people working at ISAS have told me that the net is much less useful in the afternoon and evening (in Japan) than in the early morning. Note that the ISAS ETE delay as a function of time shows little diurnal variation, though the ISAS reachability shows some. Is this relatively minor diurnal variation in ETE reachability the source of the variation that experienced users tell me about? If not, how does one better quantify the problem?

    For the record, the SAG ETE delay and SAG reachability show a plausible diurnal variation.

    Thanks to Andy Phillips for the shell script and the IDL data analysis software, Michael Foley for setting it up at MSU, Miura san for the information about the net in Japan, Joe Khan for help with the figures, and Hugh Hudson for help with the gif figures and the html.