i have a pair of cico 6504 running vss and layer3
a pair of n5k vpc to the 6504 and run as layer2 switch
one n2k and one cisco2960 dual-home upstream switch to the n5k with vpc.
first,let my laptop connect to the 2960 and ping the svi interface on 6504,reload one n5k,i got "Lost = 3 (6% loss)" ping timeout when it's up.
now let my laptop connect to the n2k,the same testing but no ping timeout
the delay restore time is 120s on n5k vpc domain
when the delay restore time modified to 90s or 30s,packets loss happened on n2k but nothings changed for 2960. like port-channel was up very early on 2960 during n5k start
but packets can not be forwarded. how to control port-channel forward packets to another n5k until reloading finished completely
I can't remember the load balancing algorithm for the 2960s. Can you change it to be IP base and try again?
port-channel load-balance src-dst-ip
When you reload the n5k the portchannel should continue forwarding packets using the other link that connects to the other n5k even if the first link is up.
as you've said.i changed load balancing algorithm to src-dst-ip for 2960,and packets loss happened with some of the pc host,the algorithm itself have on concept of the time n5k can start forwarding
i think it's incompatibility between n5k and old switch
but i can't get 2960 connect to 6504 directly,somebody told me that change the spanning tree to mst would reduce packets loss to the minimum and i can't do the mst testing in production network
You cannot on which N5k the packets would be sent to but it should work whichever N5k gets the packet. We had one
known issue due to which the vPC secondary switch would not send frames over peer-link during delay-restore timer.
Which NX-OS are you running? You might want to test with 5.0(3)N2(2a) which has fix for
Broadcasts sent from vPC secondary to the peer-link will not transmit on the vPC primary destination vPC. This is usually seen when the secondary switch has orphan ports or L3 interfaces that are up while the vPCs are suspended.
The observed impact is noticed while vPC delay restore is in effect. Traffic must be successfully forwarded across the peer-link to transmit out the primary's vPCs. Since this is not happening properly, the delay restore timer affects the length of the outage.
This problem happens specifically on Nexus 5548 and 5596, but could not occur on Nexus 5010 or 5020.
Reduce the delay restore timer to minimize impact, but tune to give enough time for routing protocols to converge:
vpc domain [id]
delay restore 10
n5k version is 5.0(3)N2(1) for us
but what i don't understand is "broadcasts sent from vPC secondary" in my opinion,vpc "secondary" only on control layer,forwarding occurs at the same time between primary and secondary n5k.
actually,i want to know the difference between the port-channel on n2k and it's on 2960.