Solved: Campus: GLBP and RSTP Root

Sonugnair_2 · ‎10-03-2010

Hi,

Assume that there are 30 vlans, two 6500 switches and 30 edge locations. The edge locations are duala homed to both 6500s. Th SVIs for all vlans are on the 6500 and GLBP is the FHRP. RSTP is used to prevent bridging loops. Even Vlans have GLBP higer priority on Core 2 and Core 2 is also root. Odd vlans have higher GLBP priority on Core 1 and is also root for RSTP.

In this scenario, i have a doubt. Assume an edge location vlan 5. Now vlan 5 could be blocked in uplink to Core 2 and active on Uplink to Core 1. Core 1 would be the AVG for vlann 5 as it is having higher GLBP priority. Now since by default round robin is used, assume that a Host B gets reply to ARP of VIP to be MAC2, which is Core2. But frame to Core 2 can be forwarded only via uplink to Core1, because uplink 2 blocks vlan 5. So frame destimed for Core2 reaches it only via Core 1 and a frame has to reach Core 1 because GLBP has told host B that default gateway MAC is MAC2 owned by Core 2. This would mean that there would be extra traffic on core to core etherchannel

I hope i have explained the scenario. If what is have written above is correct , then :-

1) If using L2 uplink,It is a bad idea to mix glbp and rstp in campus
2) If using L2 uplink,It is better go for hsrp and rstp in campus
3) Best idea is to go for routed uplinks.

Is the above correct

And if i have used the above scenario, would it be classified as a bad design in an network audit?

I would like to know your inputs on this.

Thanks in advance

Jon Marshall · ‎10-04-2010

Sonugnair wrote:

I hope i have explained the scenario. If what is have written above is correct , then :-

1) If using L2 uplink,It is a bad idea to mix glbp and rstp in campus
2) If using L2 uplink,It is better go for hsrp and rstp in campus
3) Best idea is to go for routed uplinks.

Is the above correct

And if i have used the above scenario, would it be classified as a bad design in an network audit?

I would like to know your inputs on this.

Thanks in advance

You are absolutely correct. Using GLBP in the scenario you describe would indeed mean more traffic going across the L2 etherchannel trunk between your core switches. And to a large extent you are losing the benefit of GLBP because the bottleneck is more likely to be the non-blocked uplink from the access-layer switch than the core switches. This doesn't mean it's bad design as such but it is you are not utilising fully the benefits of GLBP.

As for solutions, well it depends on whether you are comfortable with all the extra traffic on the core interconnect. If you are not then RSTP + HSRP would be a solution but i'm assuming your interconnect is an etherchannel at least and is not overloaded.

2 solutions spring to mind -

1) routed links as you say and then there is no need for GLBP on the core. This works well in a campus environment but it does restrict your options ie. you cannot have the same vlan across multiple, non-connected access-layer switches

2) the 2nd solution is actually a perfect fit for GLBP and that is to have a L3 link between your 2 core switches and not a L2 trunk. This way there is no STP loop and both uplinks from each access-layer switch can forward so using GLBP in this scenario makes a lot of sense. L2 adjancency between the 2 core switches for each access-layer switch is provided by the access-layer switch itself so GLBP etc. will still work.

Having said that. both 1) and 2) require a fair bit of reconfiguration but more importantly some redesign of your network so it really depends on how much you want to change it.

Jon

View solution in original post

Jon Marshall · ‎10-04-2010

Sonugnair wrote:

I hope i have explained the scenario. If what is have written above is correct , then :-

1) If using L2 uplink,It is a bad idea to mix glbp and rstp in campus
2) If using L2 uplink,It is better go for hsrp and rstp in campus
3) Best idea is to go for routed uplinks.

Is the above correct

And if i have used the above scenario, would it be classified as a bad design in an network audit?

I would like to know your inputs on this.

Thanks in advance

You are absolutely correct. Using GLBP in the scenario you describe would indeed mean more traffic going across the L2 etherchannel trunk between your core switches. And to a large extent you are losing the benefit of GLBP because the bottleneck is more likely to be the non-blocked uplink from the access-layer switch than the core switches. This doesn't mean it's bad design as such but it is you are not utilising fully the benefits of GLBP.

As for solutions, well it depends on whether you are comfortable with all the extra traffic on the core interconnect. If you are not then RSTP + HSRP would be a solution but i'm assuming your interconnect is an etherchannel at least and is not overloaded.

2 solutions spring to mind -

1) routed links as you say and then there is no need for GLBP on the core. This works well in a campus environment but it does restrict your options ie. you cannot have the same vlan across multiple, non-connected access-layer switches

2) the 2nd solution is actually a perfect fit for GLBP and that is to have a L3 link between your 2 core switches and not a L2 trunk. This way there is no STP loop and both uplinks from each access-layer switch can forward so using GLBP in this scenario makes a lot of sense. L2 adjancency between the 2 core switches for each access-layer switch is provided by the access-layer switch itself so GLBP etc. will still work.

Having said that. both 1) and 2) require a fair bit of reconfiguration but more importantly some redesign of your network so it really depends on how much you want to change it.

Jon

Sonugnair_2 · ‎10-04-2010

Thanks a lot for your reply!!!!