new idea for lunar - auto selecting download mirror using country codes
sofar at lunar-linux.org
Fri Jul 22 21:55:24 UTC 2005
Jon South wrote:
> Eric Sandall wrote:
>>Would the use of netselect be a better idea?
> In principle that is a neat utility, however seeing that it doesnt
> actually test any kind of actual connection, connection speed, and uses
> protocols that are likely to be marked as low priority, I wouldn't trust
> the results to be very accurate.
> For example, everyone knows there are several sourceforge mirrors. Many
> of these mirrors are fairly reliable, but vary in speed -- and some even
> limit downloading speeds. the OSDN mirror is a prime example of this,
> due to the fact that OSDN already hosts all of SF.net, it stands to
> reason they would limit the speed of their primary mirror favoring
> better speeds from dedicated mirrors.
> The OSDN mirror usually limits my downloads between 20 to 30k/sec,
> whereas the switch mirror usually spits out my maximum download; however
> take a look at this result:
> # netselect -v switch.dl.sourceforge.net osdn.dl.sourceforge.net
> Running netselect to choose 1 out of 2 addresses.
> 264 osdn.dl.sourceforge.net
> There may however, be a compromise at the expense of a little time (both
> in development and user patience) -- perhaps a script to monitor
> download speeds and if it drops below a certain point for a specified
> time, switch to the next mirror. However, I dont know if this is
> feasibly possible without some serious scripting time, and probably
> large changes in the core. And not many devs have that kind of time at
> the moment.
you're just probably close to osdn's mirror site, since I get
continuously the shortest route (ping time based) to a mirror if I feed
the entire SFORGE_URL lists servers to netselect. (same for gnu)
unfortunately my "relaxed" icmp/udp rules already block netselect's
packet method, so much for being usefull in general situations! I'd much
rather prefer a HTTP ping for this.
I'd consider adding netselect as a way to provide mirror-picking for
users who have no clue, but we really should prune the bad mirrors from
the list instead, which is far more effective
only then will a tool like netselect become usefull, and only then will
it actually spread out bandwidth usage across mirrors. Still, might be
an idea to add netselect to moonbase so it can be used for experimenting.
More information about the Lunar