Here is the present situation:
6509A --- 4500
||``````````````| where ` = blank spaces
Link to 6509B is blocked by STP. Link is L2. HSRP runs between 6509s with L2-etherchannel.
- I am wondering if L3 links between 4500-6509 and 6509-6509 is better ?? I mean by defining:
a. connected Gig ports of 4500 and 6509s as L3 ports with IP addresses in the same subnet
b. at 4500, define a default gateway to 6509 hsrp address
c. between 6509s defining a L3 PAgP etherchannel.
Whats your advice ?
I propose to have layer 2 link from each 4500 to both the cat6k switch and run HSRP on cat6k switches and configure the links between 4k and 6k switch as trunk links.
Incase if your topology and network design needs a link between both the cat6k switch I propose to be a layer etherchannel to avoid any STP loop incase if you confiure it as layer 2 etherchannel.
If I understand correctly, you recommend L3 etherchannel between 6509, and L2 between 4500-6509s.
- In case of L2 link (trunk) between 4500-6509a and 4500-6509b, one of the link is blocked by STP so only one link is in use at a time.
Question: If these are L3, both can be in use at the same time. All connected ports on 4500 and 6509s will be in the same subnet. Is that recommended ?
Also, presently L2 link is trunk carrying several vlans so I'll have to define L3 subinterfaces for each vlan. Whats your opinion ?
The links between the 6509s should be L2 etherchannel and the links to the 4506s can be either L2 or L3.
If the 4506 supports L3 then you can remove STP from the uplinks and if you use OSPF or EIGRP then the routing protocol will load balance across the uplinks for you. Plus failover and recovery will be very fast.
I also wanted to add that if L3 uplinks are used from the 4500 to the 6500s, each uplink needs to be in a separate subnet. If both uplinks are in the same subnet then you are right back to where you started in that the uplinks are now L2 uplinks and not L3 bringing STP back into the picture. The L3 uplinks are ussually created with /30 subnets and the users default gateway is on the 4500.
The pros are that the L3 etherchannel will not forward any broadcasts from the subnets/VLANs on each of the 6500s. Also, you remove STP from the link when configured as an L3 etherchannel which can provide greater network stability not only on the 6500s, but if these are core type switches that aggregate many other switches, then the rest of the network will gain stability as well.
The cons are that a VLAN and/or subnet that exists on one 6500 cannot exist on the other 6500. This is probably the biggest con. If these switches have servers directly connected, then these servers cannot use redundant configuration on their NICs. Redundant NIC configuration on servers is becoming very commom these days. As a consulting engineer I have not worked in one environment where I have been able to implement true L3 links between core 6500s because of redundancy for something or other.
Hi Mark, Thanks for the reply. If L3-etherchannels are defined between 6509s, a VLAN that exists on one 6509 can't exist in the other 6509. In that case how will HSRP be set up ??
Presently, 6509--4500 link is L2, and VLANs are defined at 6509 and trunks conects both switches. If 4500--6509 is L3, same vlans can't be used at 4500 that are defined at 6509 .. Will break the whole vLAN structure and will become more like a WAN. Whats the pros and cons of a L3 between 6509-4509?? One advantage with L3 is both links can use eigrp load balancing but with L2 also we could divide vlans on trunks to spread traffic on the two trunks.
Load balancing an L2 uplink the way you have described is a good practice that I have implemented many times.
What you mention about doing an L3 link between the 4500 and 6500 is also accurate as it will "break" the VLANs. What I do when I use L3 links to the 4500 (or any L3 switch) is I move the default gateway to the 4500 and no, those VLANs cannot exist else where in the network unless the other switches have a L2 link to the 4500.
The pro to doing this that user PCs running windows generate a lot of broadcasts and these broadcasts will not get past the 4500 keeping the uplinks lean and mean. Also, as you mentioned, OSPF or EIGRP will load balance the uplinks and when the uplinks are configured this way there is no STP on the uplinks removing STP reconvergence times providing faster failover and recovery.
Is one better than the other? They both work very well but utlimately it's your network so it's your call.