1) How ICMP is handled on routers
ICMP actions are not handled in the "fast path" in most routers and
consequently, latency numbers will be skewed waiting for the router
processor to get to them and drop numbers will be larger because they
will get dropped more frequently. MCI recently sued one of the internet
weather sites over this methodology.
2) Effects of asymmetry
The reverse paths seen from a given hop along the way to a destination
are frequently not the same as the reverse paths seen from the
destination. This is what gives rise to the strange behavior where
traceroute latency numbers go down as you get closer to the destination
(Seemingly MORE hops away).
Again, this is doable... it should just be done with eyes open.
- Stefan
> -----Original Message-----
> From: tom@emigrant [SMTP:tom@emigrant]
> Sent: Monday, July 27, 1998 12:39 PM
> To: syn@cs
> Subject: eliminating the shirikan (sp?)
>
> Although it just occurred to me, this seems so obvious it probably
> has been suggested already; if so, I apologize for repeating it.
>
> Because we are using traceroutes, rather than pings, we have
> measured round trips to interior routers, not just end points.
> So one way to eliminate the shirikan or hook routing (where
> the Detour route goes over the same link twice) would be to see if
> we get better Detour performance by using intermediate routers as
> transit points, rather than only end nodes as we do now.
>
> tom