|
North American Network Operators Group Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical Trends and Comparisons in Global BGP Data
"There are three kinds of lies; lies, damned lies and statistics" - Benjami Disraeli Because I was curious, I merged 2, then 3 of these bgp graphs. This is done using a graphics tool and is not at all alleged to be accurate, but it should be illustrative. http://www.routingloop.com/share/2bgps.gif - geoff's and tbates' http://www.routingloop.com/share/3bgps.gif - geoff's, tbates', and jhma's http://www.routingloop.com/share/overlay.gif - same as 3bgps.gif with 2 extrapolated trends Conclusions drawn (none of them revelatory): Certainly the data is dissimilar, w/ geoff's generally higher, undoubtedly due to the reasoning below: > So whats going on? Inside AS1221 there is a fair number of local routes > (about 22,000 of them). Over the past three months AS1221 been removing > noise components from the external view of AS1221 (such as removing > asymmetric satellite services using BGP routing), and the view on these web > .... The data from different sources is quite different. My experiences at a large ISP subsidiary of a large telco, upon considering merging w/ another large telco, caused me to do tremendous analysis of BGP information, such as comparing 'size' in several metrices, such as routes, traffic, address_space, etc.. What I found (in this past life) was that correlating the data from different sources was particulary difficult, if not impossible, and that most all views into the global routing table were indeed different, as Geoff states below: > My personal take on a bottom line: every view of the BGP table is relative, > and changing local circumstances as well as changing global circumstances > generate changes in the local perspective of the BGP table. Its sometimes a It does seem that we've seen a bit of a slow down in routing table growth. Someone with more time should take a look at breaking down the implied curves over given periods. It sure looks like we were on a slow exponential curve from ~1998 to 3Q2000. 3Q2000->Now looks like more of a linear growth. I suspect if one asked these 3 folks for the tabular data, a wiz with something like mathematica could do some really nifty analysis. -alan legend -- cyan = ~jhma's work red = geoff's work grey = tbates' work purple = extrapolated trends
|
