I am currently trying to design an infrastructure for a client. Bear in mind that this is my first design, and that I am learning a lot as I go (Working on CCNP R&S - I just passed Switch last monday).
After suggestions from a more experienced network engineer, I have opted for a full Layer 3 design - all the way to the access layer switches (client has the budget, and will eventually move to VoIP and interactive video, so I prefered L3 over L2 design for convergence/failover etc..).
The client has 2 sites, linked together with a LAN extension. The whole network is Cisco equipment, EXCEPT one, a BNT L3 switch in an IBM BladeCenter. This is a completely new infrastructure, will be built in parallel to the actual network.
The BNT made me reconsider my plan to use EIGRP for the design, since it's not supported by non Cisco equipment. However, I could use static routes between the BNT and the 4900M and keep the EIGRP topology. There should be one or two VLANs max on the blade center (production servers), so I wouldn't have too many routes to manage. My initial thought was to do a /30 subnet between each L3 switch in a point to point configuration.
I used ALS03 and ALS04 because the 4900M only has 10GigE ports, and I wanted to provide redundancy for those links. Also, ALS03 will have VLANs for the DMZ. ALS04 has the WAN links, network appliances, etc...
1. Is the EIGRP/static routes solution a good one? or OSPF network wide would be better?
2. Is the point-to-point addressing plan a good way of doing it? or a large broadcast subnet including all the router interfaces would be better?
3. I need the internet access to go through the local route for both sites, with failover to the alternate site. How would I implement that? (2nd site on the diagram is the disaster recovery site).
Please share your thoughts on my design, your input and suggestions are more than welcome! I am confident I can deliver a working solution but doing this on my own is kind of overwhelming!
1. define better. Either will work, If you think there may be some Non cisco devices that do layer 3 functions in the future, then go with ospf.
2. this question makes no sense to me.
3. There are a variety of ways to accomplish this. I would focus first on your internal infrastructure addressing and routing.
Other than providing a link to the metro swich, what other function does ALS04 provide?
Regarding 1: the company intends to stay Cisco as much as possible, there was no viable option available on the bladecenter thus the exception. So EIGRP seems to be the better choice.
2. lol Ok... Is point to point adressing in /30 subnets the only option in a topology like this? Between the access layer switches and the core/distribution?
3. Ok. Just wanted to make sure this was a viable solution. I just don't want to have to change some design choices because they would block that from working.
ALS04 will be used to provide connectivity to appliances, servers that are not virtualized, vmware hosts etc...
Thanks for your time!
1. either one will work just fine. pay attention to addressing and summarizing.
2. You can try a /31 with the right level of code. but if you are ptp why on earth would you allocate a larger subnet ?
3. from a purists point of view, ALS04 should hang off the core, and not be directly connected to the metro switch.
Metro can connect to the core. \
figure out your addressing plan, keeping in mind to summarize your access & server swich blocks to the core
your remote branche should also advertise a summary back to the hq.....
For the internet you could concevably use static routes, with hq having a better administrative difference than the remote path.
2. I thought about that while studying for OSPF where OSPF over NBMA network can use a single subnet...
3. The metro switch the client has cannot connect to the 4900M, they are all configured as 10GE ports. Plus, if I did that, if one of the 4900M failed, I would lose the metro link. I know, the ALS04 is now my single point of failure... Just moved it away from the core, but it's easier to replace / reconfigure if something happens...
Am I better off making the two links between the core/distribution switches, an etherchannel? or simply using two standalone links with equal cost route load balancing?
too bad you have a fixed function core. flexibility would be nice.
Myt own preference for the core to core is a routed ether channel.
here is a link for all the cisco design guides. there is a campus one in there someplace.
What do you mean by fixed function core? What else would you have used? The Nexus series? We looked at the 5548 but being that this company is fairly small (1 sysadmin) and he's not familiar with NX-OS, we didn't want to add extra complexity to the infrastructure.
There is going to be a NetApp storage solution attached directly to the BNT on the BladeCenter, so we didn't need the core switches for the storage.
I have done a ton of reading but i have a hard time finding documents that would apply to my current design.
all 10GE ports..... I do think you can configure them for different speeds. I might have considered a 4503 just for flexibility.
Well the 4900m uses X2 connectors. I could probably get a Twingig adapter and uplink @ 1GigE...
Thanks for your input, much appreciated. So much to learn A long way from CCIE but I'm working at it!