On Mon, 19 Dec 2011 10:27:03 -0500 seth vidal <[email protected]> wrote:
> On Mon, 19 Dec 2011 13:51:02 +0100 > Zdeněk Pavlas <[email protected]> wrote: > > > The fastestmirror plugin tries to find the fastest mirror > > by timing a null TCP connect(), so usually 2x RTT. > > > > However, the value measured includes host DNS lookup overhead, > > which is completely unrelated. So, perform it outside > > the connect() statement. > > > > Applying minor patches to fastestmirror doesn't fix the underlying > problem that fastestmirror doesn't measure what you actually care > about > - transfer of files - not connect() time. I must disagree. This patch fixes fastestmirror so it actually uses mirror server with lowest latency - this fixes issues where random server from same domain (same dns latency) is selected. So with this patch fastestmirror selects always the server I have behind gigabit ethernet and not the other server which is behind dsl link. This is even more obvious because there is very little difference between latencies of those two sites, dns latency is much longer than slower network path latency... I'd ACK this patch until there is better implementation - this is really worth fixing. -- Tuomo Soini <[email protected]> Foobar Linux services +358 40 5240030 Foobar Oy <http://foobar.fi/> _______________________________________________ Yum-devel mailing list [email protected] http://lists.baseurl.org/mailman/listinfo/yum-devel
