John Ricketts kirjoitti: > I am curious about how the mirrors are load balanced... > > For example, when I do my own "yum update" on the very machine that > hosts the mirror, it determines that my localhost machine isn't the > fastest site. In fact, I don't get as much traffic as I had anticipated > in general. > > If this is explained elsewhere, please excuse my not RTFM'ing. For those using the yum-fastestmirror plugin, the fastest mirror is determined by the time it takes to resolve the mirror name and to connect to the server's TCP port. In particular, it does not measure the bandwidth from the mirror. This method isn't perfect, but it does usually pick one of the nearest mirrors. Sometimes there are some oddities, as you have noticed. Try sort -k 2 /var/cache/yum/x86_64/6/timedhosts.txt or sort -k 2 /var/cache/yum/timedhosts.txt , those commands will list the connection times to the mirrors yum knows about, sorted by measured connection time. http://miuku.net/tmp/testconnectiontime.py.txt can be used for running connection time benchmarks on your own. That script has been extracted from yum-fastestmirror.