3

I am in the planning phase of re-cabling our 5 racks in one of our offices and I would like to ask for some guidance on how you go, or would go about managing cables that goes between racks. In our situation we have 5 racks where the furthest to the right is our main patch panel for 300 floor ports. The rack next to it is our main comms rack where main switches and ISP routers are located. the other 3 racks next to the comms rack then all need to connect back to the main comms rack.

I am not sure if a 48 port patch panel in each rack would be any good for this scenario? mainly because i am not sure this can be linked back to the main switch with only 1 cable.

Would a 48 port switch in each rack be better as you can uplink those back to main switch?

Or should we just run cables between racks back to the main switch?

Hope someone can offer some guidiance.

4
  • 4
    Pictures would be a big help in understanding your situation, and there's nothing quite like before and after cabling porn. Oct 16, 2012 at 21:07
  • Picture will be uploaded tomorrow once i get back to work. good idea. thank you for the feedback
    – RippieUK
    Oct 16, 2012 at 21:09
  • You'll need to provide as much info as possible, what do you need at a minimum, your budget, what bandwidth do you require. how many ports do you need, are you teaming/aggregating ports for high availability. Any problems with fiber.
    – daxroc
    Oct 16, 2012 at 21:09
  • You don't mention what's in the three racks - is it 1U servers? Storage nodes? Blades? What kind of cabling density per server? If it's 3 racks of 6 servers apiece then cross-patching using appropriately laid out cable trays isn't a big deal and almost certainly doesn't call for 3 48-port switches. If you're running pass-through modules on blade enclosures and can run 4-5 enclosures per rack then the consideration is quite different.
    – rnxrx
    Oct 17, 2012 at 0:53

2 Answers 2

5

My rule of thumb based on years of building server rooms: Minimize cross-rack cabling as much as possible.

The 300 port rack for the edge ports is far from full so you can place the edge-switches in the same rack. This keeps most of the cabling in the same rack.

The 3 racks to the left: I presume those hold your servers. Fit a cheap gigabit switch in each. Use 2 if you have servers requiring redundant links. HP ProCurves or Dells would fit the bill nicely. If you use 1 don't forget to cable it redundantly to your cores. If you use 2 redundant switches they only need a single uplink (to different cores) each.

Link all 4 racks with copper to your cores in the comms rack. Distances don't warrant fiber and copper is still a lot cheaper. Use multiple 1 GB copper links with aggregation to increase bandwidth if needed. 10 Gbs copper might be an option too depending on your cores/switches.
If you have servers that require a 1-on-1 link with a core or ISP equipment then just run an extra UTP cable. That is not going to kill you.

If your cores are big blade-switches that also carry the edge ports for the rightmost rack seriously consider moving those cores into that rack (space permitting). No one in his right mind wants to cable 300 UTP cables between racks.

You may have to spend some money on extra switches, but that will pay for itself by minimizing future support hassles.

1
  • Learned this the hard way doing physical datacenter collapses (move multiple server rooms down to one big facility). started out doing a 48-port patch panel in each rack tied to a matching patch panel in a central coms rack. It is so much easier to drop a 48-port switch in the top of each rack and run a couple cables back to a central location. just remember to label the ports based on the rack for an easier time later on.
    – TopHat
    Oct 17, 2012 at 4:12
4

Please don´t view this as an answere because this is your personal choice depending on your circumstances and budget. I can describe you what we did some month before in a structure in a (I think) comparable size with very low budget.

We have to manage around 5 distinct networks and had the same number of ethernet links between the racks distibuted in our buildings which became worse and worse to maintenance. We decided to replace 3 old switches to have all of the same brand (Dell) and type, to introduce VLANs and to use optical connections for the longer dinstances. Fiber connections are very cheap nowdays and very easy to install.

The patch panel racks were connected by 2 optical links (each) to the two main switches which themselves are connected to each other by 4 copper ports (trunk). In each server rack we now have one 48 port switch and at the patch panels there are 3x 48 ports (we have two patch panel racks) and 1x 24 ports for a smaller department. The patch panel switches only have one uplink to the main switches and are interconnected by copper ports (using port trunking there too).

Now we are in the comfortable position to be able to define for each port on each switch in each rack which network it belongs to (VLAN) and don´t have to change physical connections (most of the time). Since we use virtualization it even becomes more flexible because for the VM-hosts we just need 1 physical links to attach the VMs to all networks. There is still work to do (migrate the firewalls to use VLANs instead of physical ports) but we are really happy with this solution so far and it was really cheap.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .