Re: random host
| From: | Hans Zaunere | Date: | Tue, 20 May 2003 12:38:38 +0000 |
| Subject: | Re: random host | ||
| References: | 1 | Groups: | php.mirrors |
| Request: | Send a blank email to php-mirrors+get-17848@lists.php.net to get a copy of this message | ||
--- Gabor Hojtsy <gabor@hojtsy.hu> wrote:
> >>But if we have a select few mirrors, having this automatically work with
> >>the user's country will be hard. There is no data available whether a
> >>German mirror would be closer to me or a British one. Germany is closer
> >>geographically, but I may have a much better connection to Great Britain.
> >>
> >>If we would be able to trust all mirrors then it would be much easier to
> >>make this thing "close mirror" friendly. But what if there will be no
> >>trusted mirrors in Hungary? How can some code detect which mirror has
> >>better connection the network I am connected to?
> >
> > Geographic location, or closeness, isn't neseccarily related to Internet
> > 'closeness'. Typically, the closer a server is to a backbone, that is a
> > major upstream provider, the closer that server is to all other servers.
> > Trying to detect what country a server is in isn't as useful as knowing
> how
> > well connected the server is; what counts are hops, and after that it's
> the
> > bit pipe.
>
> If a mirror is in the same country then me, that it is probably very
> fast, as AFAIK every country or region has it's own Internet Excange,
> where the local providers are interconnected to exchange local traffic.
> At least we have this :)
>
> So having the php.net backbone some user's can get immediately to a
> close mirror (a mirror in their country) if such a mirror exists on the
> backbone. Other countries can only get a random mirror.
>
> Note that I am just playing with the ideas to get this random stuff a
> bit more close to user expectations.
Understandable, but if a user can tell the difference between a host in their
country, and one at a major peer-point in Europe I'd be very surprised :)
And the same goes for states here in the US. For instance, it's often faster
for me to hit a UK/German PHP mirror from one of my locations here in NY,
even though there's a mirror that's geographically much closer. Why?
Because the UK/German mirrors are very well connected, where as the mirror
that is actually geographically closer to me has connectivity/resource
problems.
In a previous message, James suggested that we take the top ten mirrors and
begin to implement the random.php.net round-robin system. I'm all for it,
but am not sure how we'd find the top ten mirrors. If the mirror list is
available in a nice plain-text format (*not* XML :) I'd be happy to run it
through some network metric software to help determine which are good
canidates for the top ten.
Best,
H