Description of problem: Looking at the web server logs, and at tcpdump packet dumps, it seems that during the main RPM-by-RPM installation phase, anaconda is requesting each RPM *twice*. It doesn't use HEAD or get a partial copy - it GETs the whole darned RPM each time.
It should be just doing byte-ranges to get the header for the first stage. What web server are you using?
The web server runs apache 2.0.53-3.4.legacy with a customized configuration. The byte range makes sense, however in the tcpdump packets, I recall no sign of the Range:-related headers. In case a web server doesn't support byte ranges (for whatever reason), do you think it is reasonable for anaconda/urlgrabber to request two copies?
Can you attach some of your web server logs or tcpdump output?
The web server logs simply indicate consecutive identical downloads: dhcp7.elastic.org [192.168.1.17] - - [02/Mar/2006:11:52:36 -0500] "GET /%7Efche/FC5/Fedora/RPMS/xorg-x11-drv-sis-0.8.1.3-1.2.i386.rpm HTTP/1.1" 200 289210 "-" "urlgrabber/2.9.6" dhcp7.elastic.org [192.168.1.17] - - [02/Mar/2006:11:52:37 -0500] "GET /%7Efche/FC5/Fedora/RPMS/xorg-x11-drv-sis-0.8.1.3-1.2.i386.rpm HTTP/1.1" 200 289210 "-" "urlgrabber/2.9.6" Searching backward through the logs, I see when the range query was done during the planning stage: dhcp7.elastic.org [192.168.1.17] - - [02/Mar/2006:10:04:22 -0500] "GET /%7Efche/FC5/Fedora/RPMS/xorg-x11-drv-sisusb-0.7.1.3-1.2.i386.rpm HTTP/1.1" 206 3558 "-" "urlgrabber/2.9.6" The same pattern appears for every RPM installed, starting libgcc-4.1-0.27.
Yep, we certainly were downloading each package twice. Fixed in Rawhide. Thanks for the report.