Duct mounted switches in fiber to the desk environments

Unanswered Question
Sep 25th, 2009

Hi all,

we are planning the installation of ~300 duct mounted switches in a building with an existing fiber to the desk cabling (62,5µm) in order to implement VoIP with PoE.

The duct mounted switches have 4 RJ45 ports, 1 RJ45-Uplink and 1 SFP-Uplink.

Unfortunately, new TP-cabling is not possible.

At the moment there are 2 Cat4000 Access-Switches with sufficient 100FX Ports in this building.

One customer request is a redundant design to the "access-layer" (absurd, I know...), that means:

Cat4000(1) <--> DMS(1) <--> DMS(2) <--> Cat4000(2)

with the part in the middle 150 times...

(Of course there is also a connection between the Cat4000s).

Now my question:

Are there any recommendations how to design something like that?

And: How many switches should be at most in one RSTP-Domain?



I have this problem too.
0 votes
  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 2.5 (2 ratings)
Christopher Sto... Sat, 09/26/2009 - 01:28

I would recommend splitting the network up into subnets to minimise the number of switches in one broadcast domain. You have a total of 4-5 access ports per switch so if you create /24 networks that would be around 50 switches per network. the more you segragate the network at layer 3 the better, but you will have to strike a balance between managability and reliability. Whatever you do, definitely use rapid spanning tree, and enable root guard on the ports going to the dms switches. I would also recommend udld as given the amount of fibre you have makes uni directional links more likely.

Rolf Fischer Sat, 09/26/2009 - 03:04

Thank you for the response.

Creating varios new /24-networks for the switch-management will be no problem. More crucial are the existing client-vlans with their /21 masks and nearly 1000 useres. The distribution-switches are L2 at the moment and I'm not sure if we will be able to change that (but I really hope so).

Next week we want to test a pair of the switches in a lab environment concerning things like PVRSTP, performance, power consuption and things like that.

Btw: I don't think that they support UDLD.

What we cannot test is the behavior of high numbers of switches and I'd like to recommend redesigning the existing network before implementing such a high number of 4port-switches which (unfortunately) have to participate in STP.

Searching for arguments I didn't find anything about limitations how many switches are acceptable in a single L2-Domain so far.

Christopher Sto... Sat, 09/26/2009 - 08:42

What model of switch are they? If they are not Cisco, the they won't support UDLD, but you'll still be able to use loopguard and rootguard on the cat 4000s.

With regards to the limit of the number of switches in a spanning tree domain, the general recommendation is as few as possible, but primariy to reduce the complexity of the spanning tree, and to prevent the impact of loops.

As a switch loop could take down the entire layer 2 domain, the smaller the domain the better!

mmacdonald70 Mon, 09/28/2009 - 05:30

Have you looked into the possibility of using a fiber to copper transceiver with POE and then using the Voice Vlan command on the 4000s?

Then you could use the existing infrastructure instead of deploying a whole bunch of new switches.

Cat4000 --> Transceiver --> Phone --> desktop

One example is here:


Rolf Fischer Tue, 09/29/2009 - 07:39

The problem is that we are responsible to deliver our LAN-service up to the swichports, which are -in this case- the RJ45 ports of the duct mounted switch. Because we have a centralized management (which means we're not on-site) we need remote-management capable devices and the design has to be redundant.

Otherwise those transivers would of course be a possible solution.

Rolf Fischer Tue, 10/06/2009 - 08:23

Finally I found an information regarding the maximal number in the Release Notes for the Catalyst 4500:

"Ensure that the sum of the logical interfaces across all instances of spanning tree for different VLANs does not exceed 3000 for the Supervisor Engine IV and 1500 for Supervisor Engine II-Plus. The sum of all logical interfaces equals the number of trunks on the switch multiplied by the number of active VLANs on the trunks, plus the number of non-trunking interfaces on the switch. (...)"

Thanks for the replies.


This Discussion