Re: replacing ICMP pinging with SONAR

Hi,

Duane Wessels wrote:
> Since I originally looked at SONAR I've found a couple reasons to
> NOT replace the 'pinger' with SONAR. One is that Squid would lose
> the ability to send source pings at ICMP packets.

Wouldn't it be the same if the SONAR server were running on the
same machine as Squid, then Squid just sends the query?

> The other reason is that SONAR only defines the message format between
> a SONAR server and client. It does not (yet) define how the metric is
> calculated.

Yeah it's still a work in progress, it seems. Looks like the simplest
metric is the RTT.

> So, in general, it is not possible to compare two SONAR metrics
> from two SONAR servers (for the same destination).

This is what I was hoping for - some way to triangulate the "location"
of a server. If you can simplify the Internet into "local" and
"foreign" components, the way I can, since the two are separated by congested
links, it would be easier. I compare the SONAR result from a local
server to a result from the US, and decide whether I should use a local
or a foreign parent.

Since on the other hand if you're within the US, it's more-or-less a flat
topology and you cannot make such an oversimplification.

This is related to the technique I proposed earlier - the use of a
"tracerouting" server that knows the IP address of the border routers.
One the traceroute goes beyond one of these, it can stop, concluding that
the desired location is foreign.

Regards,

-- 
miguel a.l. paraz  <map@iphil.net>                              +63-2-893-0850
iphil communications, makati city, philippines          <http://www.iphil.net>