MST convergence takes 5 seconds

[pit-maravich]

Hi,

I have been testing MST in my lab which has 3x Cisco Catalyst WS-C2960CG-8TC-L switches.
I configured only MST and LACP on each switch and connected them in the ring topology. Test is to see how much time it takes for convergence in such conditions.

My expectations were it should take around 2 seconds, but results show its always somewhere around 5 seconds.
I have set all three timers (forward-time, hello-time, max-age) to minimum values and even set on the backboneFast feature, but it didn't make any effect.

Are there any configuration changes I need to make to reach expected 2 seconds convergence time?
Is it possible that these small factor devices are somehow hardware limited for STP performance?

Thanks
Petar

[MHM Cisco World]

If I am right you need fast rate LACP to make SW faster shut the Port channel when it fails.

MHM

[balaji.bandi]

how is your configuration looks like. 

check the one of the good explanation from @peter

https://community.cisco.com/t5/switching/mstp-convergence-time/td-p/3380437

[pit-maravich]

Thanks guys, after further testing we managed to get 2 second convergence time, seems like some config was missing.
When using LACP and we disconnect one port channel we get 2 seconds convergence, but after some time if we reconnect the same port channel convergence is 7.7 seconds. this is the only way we dont get the ecpected 2 seconds time. 

[MHM Cisco World]

I can not decided which part of config effect this delay 

But share the config

Abd keep in mind divide mst to two half or more abd put all vlan to this instance, even if you dont use these vlan

MHM

[paul driver]

Hello

---

@pit-maravich wrote:
My expectations were it should take around 2 seconds, but results show its always somewhere around 5 seconds.

---

May I ask  what the timers you have manually set and what convergence conditions are you basing this on?
The standard mst bpdu expiry should be 3x the hello which is 2 seconds so if you are getting 5 seconds "reconvergence" then this sounds about correct

[Georg Pauwen]

Hello,

can you post the output of:

show spanning-tree mst

Maybe we can spot something...