[190556] in North American Network Operators' Group
Re: packet loss question
daemon@ATHENA.MIT.EDU (jmkeller)
Fri Jul 8 13:12:00 2016
X-Original-To: nanog@nanog.org
Date: Fri, 08 Jul 2016 11:42:00 -0400
From: jmkeller <jmkeller@houseofzen.org>
To: William Herrin <bill@herrin.us>
In-Reply-To: <CAP-guGWytSZUJ++E6UmHUv21TMJAQLq-A-RdtCuVmLG=RKUfDA@mail.gmail.com>
Cc: nanog@nanog.org
Errors-To: nanog-bounces@nanog.org
On 2016-07-07 11:53 PM, William Herrin wrote:
> On Thu, Jul 7, 2016 at 3:52 PM, Ken Chase <math@sizone.org> wrote:
>> ICMP is allowed to be dropped by intervening routers. Someone will
>> quote an RFC
>> at us shortly.
>
> Hi Ken,
>
> That's not correct. Routers might not generate an ICMP time-exceeded
> packet for every packet whose TTL reaches zero, but that's not the
> same thing. Routers dropping ICMP packets in transit would be bad.
> Protocols like TCP depend on path MTU discovery and path MTU discovery
> critically depends on ICMP.
>
> Regards,
> Bill Herrin
All we are seeing here is control plane filtering by intermediate
routers. Unless packet loss numbers start at a router and hops past it
show the same or higher losses it's not an actual issue with the
transport path at that hop. Outside of your own domain of
administrative control, you can't rely on intermediate routers
responding to ICMP (either filtered completely or rate limited
responses).
--
James
