Best practice to connect multiple network switches?

[Kdes]

Good Day

We recently bought 9 new network switches and wanted to ask, what is the best way or best practice to connect the 9 switches to each other.

My thinking was to daisy chain them then have 1 cable back to the first switch for redundancy. As per diagram A. If we go this route, do we just simply just enable spanning tree for the redundant link?

Or is best practice to connect switches as per diagram B?

The switches we have are HP 2920 48G POE+.

Thanks

Diagram below;

 

[Sinbad]

B.
Ideally you'd want 10 gig uplinks between the switches, but I suspect yours won't support that.

 

[Kdes]

Thanks Sinbad, would have been great to have 10 gig uplinks but was a bit much for us.

What are your thoughts on the way to connect the switches, what is the typical practice?
Daisy chain vs all to one ?

 

[catman37]

Option B

 

[aybbleek]

The HPE 2920 comes with a slot for a stacking module. This would be first prize if the switches are next to one another

 

[durbandave]

Option B is definitely better. Daisy chaining can cause massive problems. If Daisy chaining one of the switches in the middle could fail which will result in all switches behind that one to fail

 

[Sinbad]

Thanks Sinbad, would have been great to have 10 gig uplinks but was a bit much for us.

What are your thoughts on the way to connect the switches, what is the typical practice?
Daisy chain vs all to one ?

As I said. Option B.
Daisy chaining you create a bottleneck if one of the uplinks is heavily utilised

 

[Stratt]

You have left out the function of these switches, will they be access switches for end users? If yes then A, with a up-link to the core/distribution from at least two switches. Design B introduces a single point of failure in your network, with option A (should it have two separate up-links to the upstream devices, from at least two different switches, a single device failure will not cause a complete network failure.

 

[Nithan15]

Wont option A cause a bottleneck on the network?

 

[HunterGR]

Option B is the way to go

 

[Stratt]

Wont option A cause a bottleneck on the network?

It depends on the switches/architecture and function of the switches, but yes it has the possibility. If the switches all have 100mbit/s ports with gigabit uplinks even with the oversubscription it will be fine for users. We have well over 10 000 users in our company, and I have never seen a user average more than 10mbit/s during the day. Yes there are peaks, but that is where your QoS should kick in, even option A will have problems with traffic spikes especially if there is VOIP running over the same uplink. Ill rather take a couple seconds of congestion during the day, than risk losing 432 network points because of one switch failure. Seeing as this is POE switches the OP will most likely be using these switches for users with hard phones with a PC connected to the phone. Therefore if the distribution switch in design B fails you theoretically (at max capacity) now have 432 users (or 864 devices) offline, where it could have only been 48 (96 devices) as in option A.

This type of failure could be the difference between a whole company's users, support staff, call centre agents etc. being offline, or just 10% of the users being offline.

 

[WAslayer]

You have left out the function of these switches, will they be access switches for end users? If yes then A, with a up-link to the core/distribution from at least two switches. Design B introduces a single point of failure in your network, with option A (should it have two separate up-links to the upstream devices, from at least two different switches, a single device failure will not cause a complete network failure.

I agree.. we had it setup this way as well for our servers in the dc.. Option A with a single uplink to each of the top and bottom switch..

 

[irBosOtter]

Get one extra switch if you can and then stack it with another one. Those two will be your distribution stack. Run two up-links from the other switches, one to each of the dist stack.

 

[TedLasso]

Get one extra switch if you can and then stack it with another one. Those two will be your distribution stack. Run two up-links from the other switches, one to each of the dist stack.

This. Option B with a 'core' stack is what we have done. All access switch are uplinked twice to the core stack for redundancy purposes.

 

[TedLasso]

All switches also have two power sources so that takes care of another potential single point of failure. If an access switch dies - well tough but we have been running ours for over 500 days now nonstop. Yes, we haven't updated firmware - nothing requires us to for the time being as access to switches management very restricted.