[e2e] "PMTUD using options" draft

[Michael Welzl]

Hi,

> Even today after seeing this email and hoping that things might have 
> changed, I tried " ping" and "ping -R" (the -R  is for the IP record 
> route option) to some nodes several hops away and found significant 
> difference in the round trip delays.
> 
> Here is a sample:
> 
> xxx.yyy.zzz is a university in the US (15 hops away)
> ping -s xxx.yyy.zzz
> round-trip (ms)  min/avg/max = 60/60/61
> 
> ping -s -R xxx.yyy.zzz
> round-trip (ms)  min/avg/max = 109/156/211
> 
> 
> xxx.yyy.zzz is a company in Europe (17 hops away)
> ping -s xxx.yyy.zzz
> round-trip (ms)  min/avg/max = 156/156/157
> 
> ping -s -R xxx.yyy.zzz
> round-trip (ms)  min/avg/max = 181/260/401
> 
> Any comments?

Yes: perhaps it's due to record route itself?

You can see in our results that the overhead was generally
lower, and we did really many pings to lots of hosts, with
a very wide range of hops ...

we used "NOP" because we didn't want to estimate the actual
overhead from processing the option itself but just the
overhead that is due to an option being in there (which,
as we guessed, was the impact of slow path processing).

Trying different types of options would certainly be
interesting.

BTW, some time ago, a pointer to another NOP based
measurement study was posted here - google for

"The Need for an Alternative to IPv4-Options"

Judging from the diagrams, these results largely
concur with ours.

Cheers,
Michael