New Equipment Questions

[Patrick McHenry]

Hi,

Been tasked with coming up with a lan design to replace some old Nortel equipment. It's a medium branch office with around 100 - 125 users spread accross a couple of buildings. I'm thinking of using a collapsed core design with uplinks to 5 access switch closets to different locations. We would like to have 2, 10 gig connections to each closet - one coming from each core.

For the core I was thinking of using 2, 16 port 4500-X switches and 3750 -X switches in the access closets connecting to users. I haven't seen any design guides using the 4500-X switches for aggregation so, I wondering if this is appropriate?

I was also thinking I could use 4928 switches for the core but I would be limited to 1 gig for the uplinks to the access layer switches.  

Any thoughts?

Or limitations?

Thanks, Pat.

[Reza Sharifi]

Pat,

One other thing I forgot the mentioned in my previous posts is that the 5K series support one and 10Gig on the same device.

So, if you have a chassis with 32 ports, and if you want your uplinks to be only at one gig, than you only purchase the one gig SFPs for it.  If you decided a year from now that you want to change your uplinks from one to 10Gig, all you have to do is to buy the 10Gig SFPs and you are in business.  You can also have combination of one and 10Gig (16 port 1gig and 16 port 10gig).  Of course, the 5Ks also support FC and FCOE ports in case you need to connect your storage in the feature. With all the other platform, this not the case.

HTH

Reza

[Leo Laohoo]

I think a couple of 6500s would be a little over-kill for the site and I believe you can take advantage of VSS with the 4500-Xs. I think the Nexus 7004 might be a little too much switch as well but, is there a Nexus model that is closer in performance to the 4500-X?

First off, Cisco will soon be releasing an "update" (or upgrade) to the venerable 6500 system with something that will be able to support 100 Gbps.  How far away?  Early 2014 is what I heard. 

For a minimum level, the 6500 can very much support full VRF.  If you plan to implement 4500X and VRF, you may just be able to run VRF-Lite. 

[shillings]

4500-X series seems like a good fit for your above stated requirements, giving you plenty of future growth potential. NEXUS and 6500s will be a lot more expensive.

If you're using layer-2 links between the core/distribution and access layers, then just use FlexLinks (an active/standby pairing) until VSS is available early 2013 and avoid active STP loops from day one. You still need to leave STP running on all the switches though. FlexLink failover time is also much quicker than STP.

This Cisco webpage should be useful and the 2nd diagram illustrates 4500-X as the collapsed core/distribution layer: -http://www.cisco.com/en/US/prod/collateral/switches/ps10902/ps12332/white_paper_c11-696802.html

Note that Cisco best practice recommendations is now to apply routed links between core/distribution and access layers. You're suggesting layer-3 capable access switches, so worth considering. Just bear in mind any IOS feature set cost implications if you need multicast routing at the access layer.

[Patrick McHenry]

Layer 3 between collapsed core and access layer switches? This is the recommended practice now? I knew people were doing it but, I didn't know this is the recommended. I would actually like to do this. Do you have a doc? How is DHCP handeled, with the layer 3 access layer switch? Or, DHCP relay?

I've heard that layer 3 between collapsed core and access layer is more efficient than layer 2 - is this correct?

Thanks, Pat.

[shillings]

Best to take some of the best practices with a pinch of salt. I think it is more a case of understanding the limitations and advantages of each design and apply that knowledge to the set of requirements.

Fortunately the Campus design zone is a bit more practical. Here is the routed access doc for enterprise campus (not collapsed distribution/core I'm afraid): -

http://www.cisco.com/en/US/docs/solutions/Enterprise/Campus/routed-ex.html

Basically, your ip helper commands etc move down to the access layer. DHCP too, as the subnets are all on the local access block.

Advantages include easier fault finding (ping and traceroute benefits) and no need to understand how L2 and L3 inter-operate. Other key advantages are mitigated by MEC in the core though.