multiple spanning tree solution

[hasnaoukarim]

hello everyone, please I give you a scenario and you answer me if it will work or not thank you :).

i have 2 switche connected with 3 fastethernet (FE1, FE2 and FE3) in 2 sides

I created two vlans and I declare these 2 vlans as follows:

vlan1 and vlans 2 to fe1

vlan1 to Fe2

vlan2 to Fe3

I configure in the two switches MSTP and I map  the 2vlans (vlans1 and vlans2)  to one instance (instance 1),

knowing that the two switches are in the same regions.

there are two loops here

the first is in the vlan 1 between FE1 and FE2

the second is between the vlan 2 fe1 and Fe3

will MSTP detect a loop?  will MSTP work well ?

excuse my english because i'm french speaker  thank you

[Alessio Andreoli]

Hi Karim,

it is not clear whether two loops exist or it is just a possible scenario... If any STP version is in place, no loop will occur because this kind of protocol is spanning (literally) your layer 2 infrastructure forwarding trafic only if no loops can occur. Let's say anyway that for some reasons a loop will take place...

In this case the BPDU will let the xSTP reconverge on a loop-free layer 2 infrastructure simply recomputing the layer 2 topology/tree.

Did i answer to your question?

Alessio

PS: your english is perfect

[hasnaoukarim]

hi Alessio,

thanks for reply, and thanks to google also for my english

for loops, I mean that the loop exists in the case of deactivating STP, and i know that the PVST eliminate the loop, but

in the case of  MSTP i dont have a lab to test this solution.

concretely, i want to create a load sharing with spanning tree, having alot of vlan (That is why I have chosen the MSTP)

knowing that I can not use etherchannel for this. 

my goal is if i shutdown the fe1,  the trafic for (both vlan1 and 2)  should be shared between fe2 and fe3. 

for this scenario i test it in GNS3 with PVSTP and it worked but i dont know if it work with the MSTP.  thank you

[Alessio Andreoli]

Hi Karim,

you can't load share in this way. Regardless the spanning tree protocol you are going to chose, fe2 and 3 are in different vlan and they will never communicate unless enabling ip routing (i should say inter-vlan routing). If you have then many vlans you should implement PVSTP because even if it is true that you are going to ask to your switch some effort from a computational viewpoint, it is also true that if a loop occurs in one vlan it won't be propagated in the others.

xSTP and load-sharing anyway is a kind of contraddiction. The idea behind the STP is to BLOCK a port from forwarding and even if you will configure MSTP the idea will not really change.

I strongly suggest you to enable inter-vlan routing and to implement a dynamic protocol like EIGRP or OSPF. At that point the routing process will keep your infrastructure loop-free without blocking any port from forwarding.

j'espere d'eder

Alessio

[krahmani323 · Accepted solution]

Hello Karim and Alessio !

From the initial description there is two instances => Instance 0 (or IST) where all vlans are mapped by default & Instance 1 which vlans 1 & 2 have been mapped to.

As MSTP works on an instance basis (and not on a single vlan basis) it will compute based on instances 0 and 1.

The BPDUs will be exchanged on the three Fe interfaces (note the BDPUs for both instances will be exchanged through a “centralized” IST BPDU containing informations for instances 0 & 1).

Thus we can consider the situation will be ending with three parallel links between the two switches for the instance 1 (and instance 0/IST as well) ; as a result one link will be in FWD state and the two others in Altn BLK state. In this specific case there will not be any load-balancing.

The decision of the FWD/BLK interfaces is taken according to the famous STP rule : 1-Lowest Root Bridge ID / 2-Lowest Root Path cost / 3- Lowest Bridge ID / 4- Lower Port ID

About the failure scenarios :

Let’s imagine a default scenario on the non-root switch with fe1 as FWD and fe2 & 3 as BLK => if we cut fe1, MSTP will converge and ends with fe2 as FWD and fe3 as Altn BLK ….and so on..

In fact, MSTP instances are "internally" using the same mechanism as RSTP..  

If you want to have a load-balancing you will have to map the interesting vlans to different instances and then tune the parameters to reach the wanted situation (mst instance priority ; port-priority and so on…).

PS: In your case, the kind of situation with MSTP to avoid are well descibed here => http://www.cisco.com/en/US/tech/tk389/tk621/technologies_white_paper09186a0080094cfc.shtml#misconfigs

Let us if more information is need (I can test it quickly in lab on next monday if you wish).

Karim

[hasnaoukarim]

hi every one

Alessio, i think that you did not understand me due to my bad english

i don't ask my question correctly , i could have been asking  differently.

Karim , thanks i begin to understand, i deduced that the MSTP will detect loops based on instances not by vlans

because the bpdu dont have tag for vlan.

now i understand well the functionnement of MSTP, and now i found a solution to my probléme.

befor, i did not want to expose my problem to let my head think a little 

the pblm is:

as mentionned in my precedent comment i have 2 switch connected with 3 links but i have 300 vlans divided in 3 groups

of 100, currently each group is configured in each interfaces for divide broadcast domain and avoid loop.

knowing that i can not use the etherchannel, i want to activate load balancing and activate link redundancy in same time.

the solution to the probléme with MSTP is to create 6 instances which are mapped to 6 groups of vlans (divided equally

dependig on the traffic ) and let interfaces in trunk without vlans filtring.

the next step is to change cost for each instances in each interfaces such as traffic will be divided and shared in these three interfaces and in the case where one interface fail the traffic that was in the failed interf, it will be shared between the 2 other interafces. and so on...

Alessio, your solution of routing between vlans is interresting but i work in SP company, the topologie and the designe is different. (we don't respect the principle of subnets per vlans). i don't think that is possible with us.

if someone finds a flaw in my solution, don't wait to tell me

thank you all.

Karim