i do not understand the way path-cost are working on trunk
Cost for gigabit is 20000, right.
Cost for a trunk with two gigabit is also 20000... not good ?
Then if a trunk is loosing one link, cost do not change and MST won't choose another link (that could be better whith its two gigabit)... is that a normal behaviour ?
802.1t says the path-cost shouldn't be recalculated so it doesn't take into account aggregations of interfaces. The only time an aggregated link gets recalculated is when you change its operational status from up to down then back again. As link aggregation doesn't support ports set to different bandwidths (or it shouldn't anyway) this shouldn't make any difference anyway.
If you want a link to be suspended if it loses members, you could try using port-tracking, if your switch supports it.
Having said that - when designing link aggregations, your aggregated link should always be designed for n+1 total aggregated bandwidth, where n is your peak bandwidth requirement on the link, and +1 is one additional link on top of that requirement.
Hi Vince and thanks for all these detailed replies.
OK for n+1, this is a great "best practice" but i won't have any ports. By the way this is not a problem for us as if one link failed, the remaining one will give acceptable speed.
In fact my question is more about a little problem i have : if you look at my diagram (see attached file), in fact at the beginning i won't have 4 fibers between switch 2 and 4, but only 2 (lack of inter-room fiber). Then on each trunk there will be only one link, one port will be empty.
In the future, we will have new fiber and then i'll add second link on each trunk
Then i'd like to have MST choose fibers between switch 1 and 3. i guess i should manually set path-cost on fiber trunk on switch 2 & 4 (for exemple 40000, that is two time the cost of gigabit), i'll put it back to auto when new fiber arrives
Is that correct for you ?
Edit : well, in fact i did the test. I changed path-cost on fiber link between switch 2 and 4.
Then on each switch i did a "sh spanning-tree instance 1 & 2" and i wrote result on my diagram using colors (red=block, green=forward, blue=root). This way it is very clear which way is used for each instance and i can see all is correct.
If you don't tell me i made stupid thing by changing cost manually, i'll keep like that
