netdev
[Top] [All Lists]

Re: missing icmp errors for udp packets

To: Pekka Savola <pekkas@xxxxxxxxxx>
Subject: Re: missing icmp errors for udp packets
From: Chris Wedgwood <cw@xxxxxxxx>
Date: Wed, 1 Aug 2001 08:48:59 +1200
Cc: kuznet@xxxxxxxxxxxxx, therapy@xxxxxxxxxxxxx, netdev@xxxxxxxxxxx
In-reply-to: <Pine.LNX.4.33.0107312246060.20772-100000@netcore.fi>
References: <Pine.LNX.4.33.0107312246060.20772-100000@netcore.fi>
Sender: owner-netdev@xxxxxxxxxxx
User-agent: Mutt/1.3.18i
On Tue, Jul 31, 2001 at 10:48:37PM +0300, Pekka Savola wrote:

    Who said I was pinging Cisco routers?  If I have two servers 100
    ms off each other, I make them 'ping -f' each other.  This does
    test the infrastructure and forwarding capabilities a bit.

Well... in some instances, its a very good approxamation.

The point I was making was the ping (and ICMP echo/reply) should be
treated as indicitive only.

For this reason, I see no harm in linux dropping the odd ping packet
if it has something better to do.

    Traceroute isn't optimal as you noted, as the routers have to pull
    the packet with expiring TTL off the "fast path", and this is often
    subject to the rate-limiting considerations also.

Well... it depends on the router.  Decent modern routers use ASICs for
most of their functionality (even tricky stuff like fragmentation is
largely handled in silicon) and have processors above the ASIC
switching layer for even more complex stuff and can sustain incredibly
high rates of crud going through them.

Anyhow, I think this thread has been hammered enough so follows off
the list perhaps.




  --cw

<Prev in Thread] Current Thread [Next in Thread>