You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
With the current approach, there is a trade off concerning the mirrors_auto.num_mirrors variable:
Higher values increase the startup time of cpcache
Lower values could result in slow mirrors being selected. Also, it's not ideal if all cpcache users will only use top-ranking mirrors: that means load will be increased on already popular mirrors. Instead, load should be balanced among all mirrors.
Perhaps there is a way to do a sort of fast preselection before the actual latency tests are run. For instance, ICMP could be used to ping hosts and sort them by their round trip time.
The text was updated successfully, but these errors were encountered:
ICMP is not directly supported by Erlang and no hex.pm repository for ICMP exists, so testing the TCP connection delay would probably be more straightforward. Consider the delay for DNS resoution, which we don't want to measure. Something like:
With the current approach, there is a trade off concerning the
mirrors_auto.num_mirrors
variable:Perhaps there is a way to do a sort of fast preselection before the actual latency tests are run. For instance, ICMP could be used to ping hosts and sort them by their round trip time.
The text was updated successfully, but these errors were encountered: