Ether Channel across multiple switches

nethacker

Hello experts in the house, I am a bit confused about ether channel on cisco switches. I have a 3560 and 4 2960 switches. i want to try and connect the 2960s directly to the 3560 so that the 3560 will serve as my distribution layer switch while the 2960s will be the access switches.VLAN information will span the access layer switches from the 3560 but i don't know if the 3560 will support different etherchannel configurations. Find attached a copy of the proposed topology if it is achievable. I am thinking of connecting the 2960s together then providing just 1 ether channel uplink to the 3560 but i also want to find out if it is realistic to achieve the attached topology. Advise will be appreciated

shodown

I wouldn't' chain the 2960's together unless there was a need to. I would keep them separate VLANS as it makes live easier to manage. As for the Ether-channels you can do them up to the 3560, but if these are just PC's I wouldn't bother. Sometimes its good to have a single point of failure as it helps isolate the problem quicker.

evian

Hi nethacker

The 3560 does support multiple instances of etherchannel, so your topology is achievable.

If the access switches are within the same location another option could be to stack the switches and provide one physical connection from each switch, yet still bundle them all as one etherchannel. this would provide 4x bandwidth + redundancy.

It all depends on what you want to achieve.

evian

After having another look at the diagram I may have misunderstood. Are you trying to create 2 etherchannels from the 3560 to the access layer?

I believe one logical etherchannel can not be spread across two switches, unless stacked.

nethacker

evian wrote: »

After having another look at the diagram I may have misunderstood. Are you trying to create 2 etherchannels from the 3560 to the access layer?

I believe one logical etherchannel can not be spread across two switches, unless stacked.

^^ this is the point i have been looking for .. Thanks

The constraint is that the access switches are not in the same location. they are separated by floors and also a minimum distance of 280ft.

@shodown, I only have 8 servers and POS terminals on the network while the rest are PCs and printers.So the option of stacking seems unachievable due to the distance between the switches. Should i just trunk the 2960s then provide just 1 port channel to the router from the 3560? When you said it is better to have a single point of failure, can you give me an instance of that?

evian

Personally I would have separate vlans to segment the servers, pos systems and pcs/printers. With each vlan terminating on the 3560, letting it also handle inter-vlan routing.

Then one (or two via etherchannel if uptime is crucial) connections between 3560 and each 2960.

A single connection to the router should be sufficient if it is only used for WAN as bandwidth isn't an issue. If redundancy is key then maybe look at HSRP on the router.

nethacker

evian wrote: »

Personally I would have separate vlans to segment the servers, pos systems and pcs/printers. With each vlan terminating on the 3560, letting it also handle inter-vlan routing.

Then one (or two via etherchannel if uptime is crucial) connections between 3560 and each 2960.

A single connection to the router should be sufficient if it is only used for WAN as bandwidth isn't an issue. If redundancy is key then maybe look at HSRP on the router.

Uptime is crucial so i was wonderting if i can configure etherchannel between each 2960 & 3560 switch without stacking the 2960s

evian

Yes you can. But only with one etherchannel per physical switch.

Also ignore my comments regarding HSRP, for some reason I thought you had two routers

nethacker

evian wrote: »

Yes you can. But only with one etherchannel per physical switch.

Also ignore my comments regarding HSRP, for some reason I thought you had two routers

So i need to only configure the etherchannel as trunk on each switch right? then on the 3560 i will use different channel group number since it's locally significant? I am labbing it up right now and i will paste my configs here soon as i finish

shodown

an instance of it. I have a customer who had 2 routeres running GLBP, to a 3750 switch stack, with 3 groups of 2960's coming to them. This was designed by someone else not us. When they called us for support there were spanning tree problems, Cabling problems, Etherchannel problems. It was so much going on and data was flowing in so many directions with no diagram of how things went it became one big mess, it took me, some other engineers and cisco 3 days to solve this problem as we had to figure out what the hell was going on. What traffic should flow where and what links should be down and still have optimal traffic flow. In the end we had to do a re design cause we couldn't make since of there design and they couldn't either. With a single point of failure you can isolate problems very fast as when you go up to the access to distro or core layer you know where your pings stop or where you see a link up/down you have a problem. If each switch has redundant paths, to a redundant path, to another redundant path, you can easily have some long nights when things don't work.

nethacker

shodown wrote: »

an instance of it. I have a customer who had 2 routeres running GLBP, to a 3750 switch stack, with 3 groups of 2960's coming to them. This was designed by someone else not us. When they called us for support there were spanning tree problems, Cabling problems, Etherchannel problems. It was so much going on and data was flowing in so many directions with no diagram of how things went it became one big mess, it took me, some other engineers and cisco 3 days to solve this problem as we had to figure out what the hell was going on. What traffic should flow where and what links should be down and still have optimal traffic flow. In the end we had to do a re design cause we couldn't make since of there design and they couldn't either. With a single point of failure you can isolate problems very fast as when you go up to the access to distro or core layer you know where your pings stop or where you see a link up/down you have a problem. If each switch has redundant paths, to a redundant path, to another redundant path, you can easily have some long nights when things don't work.

+1 i really appreciate this... The above scenario is exactly what i am facing right now so i recommended that the network should be redesigned which is what i am currently trying to do.
Basically they have like 8 servers which POS terminals talk to, so i want the servers and the POS terminals to be in the same VLAN. Also they have an AD which manages Group policies. Apart from those servers, the rest of the hosts are PCs and printers only. Any advice as to which vlan i should place the AD?

shodown

a book I keep on my shelf is top down network design, it helps out when new designs are needed. If you dont' want to spend any money Cisco design zone has some good high availability designs there also.

nethacker

shodown wrote: »

a book I keep on my shelf is top down network design, it helps out when new designs are needed. If you dont' want to spend any money Cisco design zone has some good high availability designs there also.

+1. thanks
i actually read the book in and out to get what i needed for this project and now i have been able to achieve exactly what i though wasn't possible. Now my port channels are up and running from the 3560 to several 2960s(4).