Switching Advice

[Manny-Tech]

We currently are a split site school and Procurve 2910 switches everywhere. The two switches I'd class as 'core' which link the two sites are connected by a 10Gb SFP connection. The other switches are interconnected with a 10Gb CX4 module.

 

We are shortly about to have a third block added to the equation with around 200 devices connected. I have gone off to HP and pretty much started off with saying I'd like to keep a 10Gb connection between the sites and if possible between the switches. They tell me the 2920 is the successor to the 2910 and that the CX4 module is no more so it'd be an SFP module with relevant transceivers. Obviously the price that came back was an 'all-in' from HP so SFP module and transceiver for all switches plus stacking modules which was a bit of an eye-opener.

 

I have a few questions which I hope some could help me with, or at least advise.

 

 

1. Do I really need the stacking feature? I know for management purposes it's simpler but what else do I gain out of it?
   
2. HP have recommended 4x48 port and 1x24 port to cover the amount of ports - how would others interconnect these? Would you make one a core that links to our other block and then link three of them back to the core and the 5th one to another switch or would you connect 1 to 2, 2 to 3, 3 to 4 and 4 to 5?
   
3. Is it really necessary in a school environment to have a 10Gb link between 'edge' switches and the 'core'? In an ideal world if money is no object then of course but would 1Gb be sufficient?
   

 

I'll leave it at that for now because I'm sure there will be questions to arise from this.

 

Without getting too technical, I've knocked up a basic image of how our switches are currently connected. Red is 10Gb link to other site and blue is 10Gb CX4 link.

 

Edited November 10, 2016 by Manny-Tech

[FN-GM]

1. You will have much quicker speed when switching occurs between ports on different switches. Also the config is stored on every switch so if one fails recovery is quicker. They are treated as 1 switch so technologies such as STP work better. More resilient in the event of one failing. You won't need to buy 10GB uplinks to daisy chain them, data will go down the stacking cables.

 

2. I would connect them via stack. The drawing you have uploaded has a flaw. If the second switch fails it will knock the 3 under that offline. If you insist in not stacking, setup STP and have a cable from the top switch to the bottom, that way if one fails the others can continue to run.

 

3. If you have 1GB to the desktop then you should have 10GB to the core. You will have these switches for a good number of years and having 1GB links is already outdated. I have used 10GB links since 2012 (and redundant links). Remember 200 devices is quite a number and thats a lot of data to go down a 1GB connection that would usually be dedicated to 1 machine.

 

 

You shouldn't skimp on networking year. If it is a weak link it can literally cripple your system.

Edited November 10, 2016 by FN-GM

[njc235]

The 10G link between the sites (SFP+) is a great thing to have and if you have enough cores in your fibre, you could create an IRF 'ring'* utilising HP's 5800 comware switches which will add robustness and resilience to your network. The IRF will appear as a single switch from a management perspective and will continue to function in the event of a single site loss or JCB going through a fibre in a playground. *It looks like a ring but does not function in the sense of a ring and all links are active unlike in STP where one would have to be off. You end up with 20G aggregated links between your core elements.

 

Your edge stack should not be cables as you've illustrated, you woul need to link from switch 5 back to 1 to complete the stack. Stacking modules in the 2920 come with two ports so you would use 4 x 0.5m stacking cables and 21 x 3m for example. If you do not stack then you should avoid the daisy chan and patch each physical element back into switch one thus;

 

Alternatively, if budget allows, you could create an IRF'd edge stack using comware edge switches such as the 5130.

[FN-GM]

The 10G link between the sites (SFP+) is a great thing to have and if you have enough cores in your fibre, you could create an IRF 'ring'* utilising HP's 5800 comware switches which will add robustness and resilience to your network. The IRF will appear as a single switch from a management perspective and will continue to function in the event of a single site loss or JCB going through a fibre in a playground. *It looks like a ring but does not function in the sense of a ring and all links are active unlike in STP where one would have to be off. You end up with 20G aggregated links between your core elements.

[ATTACH=CONFIG]39698[/ATTACH]

 

Your edge stack should not be cables as you've illustrated, you woul need to link from switch 5 back to 1 to complete the stack. Stacking modules in the 2920 come with two ports so you would use 4 x 0.5m stacking cables and 21 x 3m for example. If you do not stack then you should avoid the daisy chan and patch each physical element back into switch one thus;[ATTACH=CONFIG]39699[/ATTACH]

 

Alternatively, if budget allows, you could create an IRF'd edge stack using comware edge switches such as the 5130.

 

Even if your second drawing is flawed. If the top switch goes off the whole lot does. Have 2 uplinks and connect them a loop. That way if one goes off the rest will still function.

Edited November 11, 2016 by FN-GM

[Manny-Tech]

1. You will have much quicker speed when switching occurs between ports on different switches. Also the config is stored on every switch so if one fails recovery is quicker. They are treated as 1 switch so technologies such as STP work better. More resilient in the event of one failing. You won't need to buy 10GB uplinks to daisy chain them, data will go down the stacking cables.

 

2. I would connect them via stack. The drawing you have uploaded has a flaw. If the second switch fails it will knock the 3 under that offline. If you insist in not stacking, setup STP and have a cable from the top switch to the bottom, that way if one fails the others can continue to run.

 

3. If you have 1GB to the desktop then you should have 10GB to the core. You will have these switches for a good number of years and having 1GB links is already outdated. I have used 10GB links since 2012 (and redundant links). Remember 200 devices is quite a number and thats a lot of data to go down a 1GB connection that would usually be dedicated to 1 machine.

 

 

You shouldn't skimp on networking year. If it is a weak link it can literally cripple your system.

 

In which case, I'm probably better of taking the stacking cables rather than the SFP modules to link the switches. It brings the cost down a little too.

 

I'm really not that advanced on switches, is STP fairly straight forward to configure? I inherited the switches we already have and I haven't had to do a great deal to them.

 

I think I will stick with the 10Gb link between switches, like you say it'd be a backwards step from the rest of the infrastructure. So would you stack like my illustration but have a cable running from top to bottom too?

[FN-GM]

STP should just need turning on and thats it. Lots of vendors turn it on by default anyway.

 

So would you stack like my illustration but have a cable running from top to bottom too?

 

I would yes. You should have enough cables with your delivery anyway to do this. The issue would be them being long enough.

[njc235]

Even if your second drawing is flawed. If the top switch goes off the whole lot does. Have 2 uplinks and connect them a loop. That way if one goes off the rest will still function.

I was merely pointing out that a daisy chain is poor design. I concur that dual links back to the core from a proper stack is the way to go if connectivity is available.

[Manny-Tech]

I've just spoken to our HP representative and their technical department recommended daisy chain from switch one to four and then four back up to one. Which in effect creates a loop but I imagine that's where STP kicks in and enables the link from four to one if one of the other links goes?

[njc235]

Do you speak to someone in HP in the UK directly?

[FN-GM]

I've just spoken to our HP representative and their technical department recommended daisy chain from switch one to four and then four back up to one. Which in effect creates a loop but I imagine that's where STP kicks in and enables the link from four to one if one of the other links goes?

 

It will disable one of the ports yes. Its quite complicated to explain how it works. But in a nut shell will choose to block the link with longest route back to the route bridge (usually the core switch).