Optical network design

[abdelnour.moussaoui]

Hello everyone,

I work as network administrator in industry company and i am about to install our local network throughout the areas of the factory. i have designed my network with all the specification needed (see the attached file). i am going to use a fiber optic connection along with Cisco switches, SFPs and SFPs+, it's my first time trying to deploy a network using  optical cable. so my point is , based on the diagram i have built can anyone suggest or analyse my network whether it is correct or not or may be any missing points or any consideration i should take while the installation. 

I am using a fiber optic cable type OM4 with 50μm of diameter.

the farthest distance that i could have is 600m.

I need a bandwidth of 10GB in some areas (as depicted) whereas the distance could reach 500m.

I have used SPF model GLC-SX-MM and SFP-10G-SR for 1Gb and 10Gb respectively.

Questions

1. Is the wavelength related to the SPF module or to the cable specifications.

2. I am using a 10 GB of bandwidth between cores switches using a patch cable (see diagram), does the short distance effect.

3. does a switch support SFP for MMF and SMF working in same switch at the same time.

any help is greatly appreciated.

[Marvin Rhoads · Accepted solution]

Good questions and nice job thinking it through ahead of time.

1. The wavelength is according to the transceiver model.

2. Between core switches you can even use the twinax copper patch cables. They are much less expensive than the SFP+ 10 Gbp transceivers. Even better for a pair of 3850-24XS like you show is to just put them in a two switch stack and then you use the StackWise 480 stacking cable between them and don't use up and 10 Gbps ports for that.

3. Yes. Just always double check the compatibility matrices when matching transceivers with switches.

1 Gbps: 

http://www.cisco.com/c/en/us/td/docs/interfaces_modules/transceiver_modules/compatibility/matrix/GE_Tx_Matrix.html

10 Gbps:

http://www.cisco.com/c/en/us/td/docs/interfaces_modules/transceiver_modules/compatibility/matrix/10GE_Tx_Matrix.html

One other suggestion I would make is to consider a 4500-X VSS pair in the core. You are about using up all the ports on the 3850s and you can get more capacity with the 4500-X. I'd recommend the 32-port WS-C4500X-32SFP+. It's about 25% more but you get 8 more ports and have the ability to add an 8-port expansion module later. It is also a non-blocking switch so you get more throughput. 

Oh - and use the SGP-10G-SR-S models for your SFP+ transceivers. They are less costly and designed for all Enterprise applications.

[abdelnour.moussaoui]

Hello,

thanks for your suggestions.

another thing is concerning me, on the SFP+ transceiver SFP-10G-SR, i doubt that the distance will exceed 400m. is it going to work? if it is no,  it should be that i have to use a single mode fiber in this area with a proper transceiver model. what do you propose.

best regards 

[Tom Randstrom]

If you are planning for 10G over 400m links, or planning on upgrading a 1G link later, you should use single mode fiber cable and an appropriate optical transceiver (SFP-10G-LR).  1G will operate to 1000m over OM4.  You could also install a hybrid (mixed) fiber cable with both OM4 & SMF in the same jacket.  This would allow you to initially use the OM4 for the 1G link and use the SMF if (when) you upgrade to 10G.   

[Marvin Rhoads · Accepted solution]

I agree with Tom.

While it's possible you may get a good link at 500m over multimode with the SFP-10-SR (or SFP-10G-SR-S), it would be best to plan for single mode over those 10 Gbps links >400m.

However that does increase cost. Realize that if your core is a stack or VSS pair, you can use both uplinks from the access switches as an Etherchannel and thus have a 2 Gbps link. Not using a stack or VSS pair means one link is always spanning-tree blocking and the other forwarding (for a given VLAN).

That would mean you could use 4 each GLC-SX-MMD vs 4 each SFP-10G-LR-S for that connection. The respective price in US$ is $500 vs. $1900 per transceiver or $5600 (list price) cost increase per 4 transceiver connection.

[abdelnour.moussaoui]

Hello sir,

your recommendations are going to be very useful for me, thanks again.

i liked the idea of VSS, it's going to work well specially as you might see that i have a huge traffic of video, by taking advantage of the two core switches it will reduce the overload.

one thing i wanna point out you have already mention and i wanna open a discussion for it if you mind,

you said with VSS "you can use both uplinks from the access switches as an Etherchannel and thus have a 2 Gbps link. Not using a stack or VSS pair means one link is always spanning-tree blocking and the other forwarding (for a given VLAN)."

actually, i was planning to avoid STP to be involved and have a stable network topology, so in the CCNP switch official cert guide book they recommend to use a unique VLAN on each access switch which means your VLANs are not permitted to span across your distribution switches and they said also which i wanna understand " each access switches has two dedicated paths into the the distribution layer, both links can be fully utilized with traffic load balanced accross them. Page 19" how is that? i didn't get it.

best regards

[Marvin Rhoads]

If the VLAN is unique on each access switch then you have to make the uplinks Layer 3 and do routing on the 2960 switches. Most designs in practice that I have seen prefer to do the routing on a core switch and only use the access switches as L2.

Often it is a higher cost license to have full L3 routing on the access switch and it also adds routing complexity in my opinion - often unnecessarily.

Even the way you mention you could still do L3 Etherchannels or ECMP (Equal Cost Multipath) over OSPF. My way you put all the VLANs on the core as well and all the VLAN gateways (L3 SVIs) are there.

Either way it will work. I like to keep it simple as possible while meeting the functional requirements.