05-29-2014 08:50 PM - edited 05-29-2014 08:51 PM
When I direct continuous ping from PC to the ICX switch mangement IP, it will show latency in some moment. At the same time I ping from my PC to anther PC thought the switch, it would not have latency.
The switch are standalone and not connect to any network. You may try it on your own ICX switch, it seem like apply to most of ICX switch.
Anyone can explain why?
Please see the screen shot on attachment.
Solved! Go to Solution.
06-13-2014 01:45 PM
first of all, pinging directly to the switch management interface is handled differently than forwarding icmp ping from one port via another port to the destination IP.
In the case of ping the management interface, you are sending packets to the switch cpu to handle and process, then reply with icmp echo reply. ICMP traffic is of lower priority, if the cpu is running some high priority process then the icmp echo reply will be delayed.
You can see that in the Round trip time in ms. Some will spike up due to cpu being busy processing something else with high priority.
In the case of pinging from your PC via the switch to another PC, the icmp ech is forwarded by the switch hardware without the need for cpu to handled, so there is no bearing on whether the switch is performing some routine background process and cause it to be busy or not. The icmp echos and icmp echo replies will be handled by hw and there is no latency ( unless there is network transit delay - but no due to cpu busy )
06-13-2014 01:48 PM
NO, the swtich do not rate limit ICMP echos and ICMP echo replies.
The latency in pinging to switch management interface is due to the fact that cpu need to process the packets and send reply, ICMP is low priority traffic compared to other control traffic, and there are routing maintenace processes that kick in from time to time, which will have an impact to inject latency in the handling of ICMP echo/echo reply for the management interface.
07-08-2014 08:10 AM - edited 07-08-2014 09:17 AM
Great to see your reply. I got the answer from you.
By the way, can we set the ICMP traffic to higher priority? Because one of my client was latency sensitive, he will not happy to see the latency on direct connect link.
07-14-2014 09:36 AM
To do this would be to de-prioritize Spanning-Tree, routing and other critical functions of the system. Explain to your client that the CPU of the router/switch is designed to perform critical network operations, just like if you asked your client every few minutes if he was there it would take away his time, it's the same theory on pinging the switch or router.
07-14-2014 09:55 AM
My problem has not been with pinging the MLX directly, it is pinging past it that is always an issue on a slightly congested pipe. Pinging the MLX directly works as expected, the pinging through it (or past) gets me lots of complaints.
07-17-2014 12:31 PM
If I did not have so many complaints about it I would not mess with it. Is there a way on a per vlan or subnet basis maybe or is it global? I do rate shape up/down speeds at the vlan level.