ICMP entering a device's control plane is usually treated as low importance traffic. Thus ping/traceroute is a poor way of measureing the performance and reachability of a router/switch, as forwarding and processing those packets will be done at a lower priority by the scheduler.
ICMP traversing a network towards an end host will in 99% of instances be treated exactly as any other traffic, and as end hosts tend to have a lot of CPU, they don't usually overly restrict processing of them.
You can see this in action by tracerouting to a linux host across the internet at low intervals, intervening hops will often drop/delay packets as their filters/prioritising kicks in showing high RTT times and packetloss, whereas the end host will show the expected RTT and zero loss.
I think a lot of people confuse these things :)
adam.
On 2014-01-28 12:25, Nikolay Shopik wrote:
Not entirely true, most networks today is FIFO, w/o any kind of QoS on it. As most unmanaged and managed switches/routers come with that configuration.
On 28.01.2014 21:01, Chris Moody wrote:
ICMP is not a very robust means of "monitoring" a device. The ICMP protocol is often lowest priority in traffic queues on routing equipment and as such will frequently just be discarded during periods of high link utilization or high traffic load. In particular, trying to "monitor" a device across the Internet, your ICMP is really best-effort. _______________________________________________ observium mailing list observium@observium.org http://postman.memetic.org/cgi-bin/mailman/listinfo/observium