I have 2 EX3400 switches that connected each other with a copper cable directly .
When I did a ping test each other via Lay2 or Lay3 interface . The latency average was around 18ms.
Also I connected a EX3400 to Cisco router (Huawei router ,Juniper EX4200 ) directly via Lay2 or Lay 3 . The latency average was around 15ms.
But when I used two EX4200 switches to do the Ping test VIA L2 & L3 . The Latency was no more than 2ms .
I don't know if this related to interface routing bridging (irb) on EX3400 .
Anyone can help me with this ? Thx
Don't ping the switches directly on the IRB interfaces. ICMP destined to the control plane on the 3400's is super low priority. Try to ping actual hosts "through" the switch and you should have much better results. It's normal behavior, and your actual latency is not that high.
Hi damenten ,thank you very much . I got the better result when i did the ping through switches .
But i still can't understand what you mean by "ICMP destined to the control plane on the 3400's is super low priority."
As usual we need to do the ping on switches to check the backbone quality . If we do ping "through swithes ",that way ,we need invest more devices . Any other ways to aviod this or any official document that i can refer to ? THANKS
When the icmp ping is destined towards the local interface of the switch, by design this packet processing will happen at the routing engine (control plane) and hence the priority of processing this kind of packet is very low on all the devices. Hence you will see latency in the ping towards an irb interface or local L3 interface.
However you will not notice that latency if the icmp is through the switch that is via data plane as it will be switches in the hardware.
Refer to the below KB for more clarity
Hi, Thanks . I reviewed the article . It seems to be a result of design of Juniper platform. but Do you know any commands or ways to slove ?
As you said this is how the platform is designed so we cannot change/modify the processing behavior using commands.