Ping return values are wrong and non linear with congestion

[dmfiore]

Hi,
I did a quick search, and didn't find anything, was wondering if there a known issue on the ping value returns being wrong ?
I find they are accurate with low values and no congestion, however when under congestion, the values being returned can be 15k to 50k of ms (ie, 15-50 seconds). For which this return is being returned with a few seconds, so that is not possible.
For example, let's say I point 10 devices to ping x100 of size 512 bytes with inter packet delay of say 3000ms. The ping times being printed to console for the 100 iterations can range up to 60000, but it reports this every 3000 ms.
Thanks for any pointers on this,
-D

[jwhui]

Can you provide a copy of the console output, including the command you used and the output you are observing? Thanks.

[dmfiore]

I'm not sure that is accurate though.
If I send x10 iterations each 1000ms apart and those x10 iterations report each took 25,000ms and the total time would be = 250 seconds for which all of this 250 seconds was reported within 10sec of realtime.

[dmfiore]

I was hoping to find thresholds of what the network can handle. And this impedes that. But certainly we should use something not based on ICMP. That said, I'm pretty sure we could change the ping API logic here for the better and perhaps reset the values and be more accurate, else we should perhaps note that ping API is only accurate if there is no congestion ?

[jwhui]

ICMPv6 Echo Request messages are sent without waiting to receive replies to prior Echo Request messages, so there is overlap in the delays for each individual request/response transaction.

The round-trip time computed with the ping command is intended to measure the latency of the network with or without congestion. This is common practice with any IP-based network.

Increased latency under load is a common issue due to network buffering/queueing - this is often referred to as bufferbloat.

[dmfiore]

Thanks for the quick responses and info, Jonathan. I still think we have a bug here.
If you send x10 packets with 1 sec packet interval(ie rate) for a total time of transmitting those commands over 10 secs.
Then how can each of those 10 replies within that 10 seconds add up to 250 seconds in total ? It only took 10 seconds for all those commands to be executed and completed with values of time.
What I can't get my head around is that if a ping reports it took 25secs, why didn't the console just wait 25secs for that value versus it being printed out within 1 sec.

[jwhui]

The round-trip time is printed whenever an ICMPv6 Echo Reply is received. The Echo Request was sent 25 seconds in the past.

Note that in your log, the first received Echo Reply had an icmp_seq value of 201 and the last received reply had an icmp_seq value of 265. As a result, the log is not showing the full 100 packets that were sent.