Re: pinger

I agree fully. Never liked the ICMP probing.

The only issue is that TCP measurements are slightly more dependent on
the server load, and cannot easily be done in parallell with the fetch
without perhaps making people even more upset..

/Henrik

Andres Kroonmaa wrote:
>
> We use pinger, to determine 2 things: RTT to the server, and hop-count.
>
> Hop-count has quite little meaning for nearly anything. Mostly people
> want to use RTT for peer selection, but this RTT has also some problems.
> quite often sites are blocking icmp ping, and may get frustrated if squid
> caches allover the world are pinging their server too often. icmp gets
> usually higher handling priority in OS'es than normal traffic.
>
> Do we really need pinger? Why can't we just measure time it takes to
> tcp-connect to the target server, (or time to first reply data from peer)
> and remember that as sort of RTT metric?
>
> We could pick neighbors by that metric, and we would be less prone to
> pick peers that are in reality overloaded, and we could also use that
> metric for decisions whether to cache replies from given server or not.
>
> just thought...
>
> ------------------------------------
> Andres Kroonmaa <andre@online.ee>
> Delfi Online
> Tel: 6501 731, Fax: 6501 708
> Pärnu mnt. 158, Tallinn,
> 11317 Estonia
Received on Thu Feb 01 2001 - 12:59:04 MST