Run lots of cables, or use a switch?

[iSteve]

Hello,

We've just had a shiny new HP ProCurve 5406zl switch installed in to our current server room comms cabinet, in to which goes our current batch of servers. It's fitted with 14 fibre ports to feed other cabinets around the school in a star topology. During the summer, our server room is moving to an adjacent room after a refurb. However the comms cabinet is staying in the old room (literally next door). It has not been budgeted, or planned that the comms cabinet, with all it's fibre will be moving, so I am debating one of the following options....

1). Run individual ethernet cables from each server, through a hole in the wall, back to the 5406zl switch.

2). Install a Procurve 1800-24G gigabit switch in each of our two server racks, make a 2Gb trunk to each switch from the 5406zl, meaning only running 4 cables which could be put in mini trunking, then having all the server connections accessible at the server racks

I feel i should plug all my important stuff, ie the servers, straight in to our core switch, but for fault finding and tidyness would prefer to go for option number 2! What sort of real world performance hit would you get using option 2 - negligible? What would you do?

[Michael]

Your first option should be absolutely fine. If you wanted to treat yourself, you could easily afford CAT6 to go through the wall and back into the switch, however good quality CAT5e is all you need

[kmount]

Option 1 with a nice massive hole would be how I'd go about it. (That's how it was done at my last place but we had a couple of metres to travel which has meant we ran out of ports in the original bundle!)

[DAckroyd]

I did the same thing just before christmas and went for Option 1 with about 25 Cat6 cables (lots of colours to make it pretty - they are also labled)

Having a couple of servers with 10 cables to each I needing to organise these somehow... unless you just want to make a big 2Gbp/s bottleneck I would go with option 1... you can box in cable fine but also having lots of cables around make it look scary to others and keeps them out of your server room... also add some fairy lights for christmas time and it will look very cool (Blue LED would be best!)

[anfy1002]

I agree use option1 But dont buy cheap non certified patch leads. For a few extra pence get NVP quoted patch leads. If your supplier looks blank when you ask the NVP then maybe you should buy them from somewhere else. I cannot believe the schools i go into to test the network where the actual network structure tests fine but the patch leads fail on test. Thus reducing the complete network throughput.

[CPLTD]

I agree use option1 But dont buy cheap non certified patch leads. For a few extra pence get NVP quoted patch leads. If your supplier looks blank when you ask the NVP then maybe you should buy them from somewhere else. I cannot believe the schools i go into to test the network where the actual network structure tests fine but the patch leads fail on test. Thus reducing the complete network throughput.

I couldnt agree more with the above the NVP should be between 70%-75% , thats why we only supply screened and tested cables with these values,

NVP for those who dont know (Nominal Velocity of Propagation. The speed a signal will travel down an electrical cable measured as a percentage of the speed of light in a vacuum)

[anfy1002]

At last a supplier who knows his onions about cabling. There is hope out there!

[iSteve]

Thanks for your responses. I'd not heard of NVP values before, but I will make sure I get decent cables in a few weeks time, and box them in....through a massive hole

[CPLTD]

Thanks for your responses. I'd not heard of NVP values before, but I will make sure I get decent cables in a few weeks time, and box them in....through a massive hole

if you pm me your requirments I'll make sure we send you over a competitive quotation!

[sdc]

We had option 1 for a while, until we ran out of CAT5 links! So we put a Gigabit switch in one of the server racks, and then used one of the CAT5 cables to link into our core cabinet (across the corridor). We have not noticed any change in network performance. In addition, we now have the benefit that, if the power fails, all of the servers can continue to communicate via their 'local' switch, thus enabling the networked UPS devices to notify all the servers. Before, the core cabinet would have powered off when the power failed, and then the UPSes couldn't tell the servers to power down!

[localzuk]

Personally, I'd go with option 1 with a twist. Rather than using patch leads from server - hole - switch. I'd be going for patch - panel - hole - panel - patch. This removes the problem that may occur of cables being stressed when being unplugged etc... For the extra few quid, i'd think it was worth it.

[iSteve]

...we now have the benefit that, if the power fails, all of the servers can continue to communicate via their 'local' switch, thus enabling the networked UPS devices to notify all the servers.

This is what we do, except when our switch was changed over this week it was re-plugged in to the mains! Next time we have a spot of downtime it'll be moved on to a UPS.

[DAckroyd]

At last a supplier who knows his onions about cabling. There is hope out there!

My network doesn't have any onions... I'll get an order in on Tuseday for that...

Just a thought though, How do I use the onions on my network? do the go well with spam?

[kmount]

We had option 1 for a while, until we ran out of CAT5 links! So we put a Gigabit switch in one of the server racks, and then used one of the CAT5 cables to link into our core cabinet (across the corridor). We have not noticed any change in network performance. In addition, we now have the benefit that, if the power fails, all of the servers can continue to communicate via their 'local' switch, thus enabling the networked UPS devices to notify all the servers. Before, the core cabinet would have powered off when the power failed, and then the UPSes couldn't tell the servers to power down!

This is fine but it does effectively mean that your servers are bottlenecked at 1Gbps uplink to the rest of the network instead of several 1Gbps connections if plugged into the core interconnect with the leaf switches.

If you specc'ed up a decent switch for this role it might not be too bad.

[Butuz]

Have exactly the same problem as you (new core switch going in to main network cab which is next door to server room).

I will be having a company in to run a 24 port patch pannel full of short links from the server room cab, into the core switch cab. It sholdn't cost more than £1000, infact hoping for it to be quite a bit less than that.

Butuz