North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

follow-on of maintenance survey

  • From: Zhen Wu
  • Date: Thu Oct 07 18:27:49 2004

Several hours ago, I post a survey request on the list. First, if it disturb some people, please accept my sincere apology.

Second, I'd like to summarize the responses, because I think it might be interesting to some people. I got about 30 responses, which is not a big size. Most people said that they do the planned maintenance during weekday midnight. Two reasons for this. First, it's the time window that will cause the least interference of network usage; Second, it's a better time to get hold of other people (customers, vendors) than weekend if necessary. For some small or critical problems, people just fix it immediately. Some people also mentioned that they would like to major maintenance during weekend.

Third, I will explain our research (routing instability) and hypothesis, which motivate this email survey. Your sharing of insight will be very helpful and appreciated. Because it's pretty long, I put the description in the end.

Finally, we still warmly welcome your survey response. Thanks!


In the academic research society, some people argue that there is correlation between network usage and Internet routing instability (of BGP).

A research done in 1996 revealed daily and weekly routing instability pattern. During the day time Internet exhibits significantly higher instability, while in the midnight we see much less instability activity. Similarly, during weekend there are much less routing instability.

We revisited this study, because we believe Internet has changed a lot since 1996. We use data from public BGP archives like Oregon RouteViews. Our new results still display pretty strong daily pattern, with higher instabilities during 6AM and 6PM (PST). However, we observe less obvious weekly pattern compared to 1996. Our visual comparisons are also confirmed by more rigorous statistics method: spectrum analysis (using Fourier Transform).

Meanwhile, Internet traffic still display strong weekly and diurnal patterns. So if we believe the correlation between network usage and routing instability, why we didn't see less instability during weekend?

That's why I send out previous survey email to see whether people do a lot of maintenance during weekend, which might contribute to some of the instabilities observed during weekend. From current survey, my hypothesis seems not to be true.

Any discussion, insight, or critique are extremely welcomed!