Re: ping and lsrping comparison

Neal Cardwell (cardwell@cs.washington.edu)
Mon, 9 Nov 1998 22:48:40 -0800 (PST)

It's interesting that the loss rate for the lsrping packets is not too
awful. This suggests maybe you could use them to help attack the big
question of where in the network packets are lost. Or, actually, where
packets are *not* lost, since we would expect LSR packets to fare worse
than regular packets. So you could send LSR packets out into thousands of
judiciously-chosen tours around the Internet, and perhaps deduce what
areas of the Internet are *not* responsible for these 5-7% and worse loss
rates that some people are seeing.

This oughta work for the interior of the net, where there are lots of
choices for paths. Of course at the edge it will be hard to tell whether
packets are being lost on the way *into* site foo, or *out* of site foo,
since there will usually be one link that you have to travel to get to and
from edge sites. I guess that's where we'd need Stefan's tool that uses
TCP state to distinguish the direction of loss...

neal

On Mon, 9 Nov 1998, Kenichi Ishikawa wrote:

> I have compared loose source route based IP packets and a normal IP
> packets.
>
> Stefan was right.
> Loose source route based packets are slow and lossy. RTT of loose
> source route based packets are scatterd. It seems that loose source
> route based packets was hugely influenced by intermediate route's
> condition.
>
> I would like to show the correlation between LSR-based packets and RTT
> of normal in terms of RTT and loss rate. But it seems to be difficult.
>
> There are some graphs.
> http://grad-pc29.cs.washington.edu/~ishi/ping_lsrping_cmp/
> Comments are welcome.
> Thanks.
>
> --
> Kenichi Ishikawa
>
>