Switch Topology

YES YES YES !!!

Why ?

MUCH easier to trouble shoot

Lower uplink utilization per 3500 switch as each one would have a dedicated link to 6500

Watch out for oversubscription of the backplane connections on the 6500, and make sure that you know which ports of the interface card have buffers in common. (Be sure you know what kind of fabric connect your supervisor / line-card combination is using, and how that fabric bandwidth is divided up between the ports.)

I can guess that's the better solution in the world but from the last sentence of Gary (...a fully circular...) I can not figure out how the connections could be done. If the 3500s are connected in circle and each one is connected to the 6500 how each 3500 can see the 6500 through another 3500?

TIA

Alex.

If a link between the 3500 and 6500 were to fail (e.g., cable gets cut or GBIC burns out) then the 6500 would still be reachable by sending the data to one of the other 3500's connected to the one with the link failure.

6500 are 3BXL's and from what you are saying I should keep both??

i.e Connect the 6500's to each 3500 using one of the GBIC ports on the

3500's and use the other GBIC ports on the 3500s to link them together with each other.

Almost like a chain of 3500's each with a spur to both of the 6500's

Assuming we have 3 x 3500's and 2 x 6500's we would have 3 x 3500's connected in series and each of those 3500's [except the middle one as not enough GIG ports] would have a connection to one of the 6500's

Would it also make sense to mix GB and FE connections so each 3500 turns up on both the 6500 switches? Both 6500's switches have multiple upstreams and peerings.

Gary

Sorry, this is beyond my experience. I would suggest that you look through Vincent C. Jones' white papers at networksolutions.com and try to find his book on high availability networking.

More connections sometimes just introduces more ways for things to break, so you need a careful plan for high availability with multiple peers -- for example, it might involve VRRP (or HSRP). It isn't just enough to have alternate topologies available: your setup has to be able to reliably detect failures (which do *not* always bring the line down), and has to reliably reconfigure around the problem.

Dual-homing access switches to dual core ssitches is a standard design practice.

I would eliminate any inter-3500 switch links in favour of using the GE's as uplinks to 6500

6500 #1 ------ 6500 #2 \\ / \\ / 3500 single switch x n

BTW can you post a show version from one of the 6500 and one of the

3500 switches

Config for inter 3500 links simple trunked ports.

Config for inter 3500 - 6500 links

6500 end interface GigabitEthernet3/18 switchport switchport trunk encapsulation dot1q switchport trunk allowed vlan a-d,f,g switchport mode trunk no ip address

3500 end interface GigabitEthernet0/1 switchport trunk encapsulation dot1q switchport mode trunk !

Why? Gary

Looking for show version from 6500 and 3500 not the trunk configs

why - because it provide the IOS version you running which translates into features that responders may want to suggest that you implement in addition to a topology change which I highly recommend you do.

3500's are 120.5.1 6500's are 122.18

Thanks Gary

So it looks like the 6500 are current

and the 3500 are just about EOL ...