In article <[EMAIL PROTECTED]>, John Labovitz <[EMAIL PROTECTED]> wrote: > here's a perl script i wrote that uses the 'ping' program to find the > 'distance' to a mirror, sorting the sites from closest to farthest.
Unfortunately raw ping times are not very well correlated to bandwidth. A better approach is to use the technique implemented by `bing' (I've packaged it), which analyzes the *difference between large and small* pings. Since ping time is the result of a raw latency component (time between first bit sent and first bit received) and a bandwidth component (time between first bit received and last bit received), by using two you can isolate the bandwidth component. Another way of looking at this is to plot an x-y graph of total ping time vs. ping packet size; you want to look at the slope, not the intercept. However, bing uses the fastest small and faster large ping it receives in an attempt to determine actual link characteristics (the size of the smallest pipe between A and B); since the important thing here is available bandwidth, you want to use the average ping times instead. It gets very tricky. A good measurement cannot be done on the fly, because it must be statistical. And finally remember that traffic patterns and even topology do change on a minute-by-minute basis. The only saving point is that a modem or 56k or ISDN line is very slow compared to a well-connected server, so any of them should be able to max out your line. So it doesn't matter all that much which mirror you use. -- Shields, CrossLink. -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED]
