Switches in chain topology for ~40 devices

Question

I am currently looking at a network topology consisting of ~40 DUTs. In this network, there is no hierarchy (like access core and distribution), and all the switches are connected in chain topology. Currently we have 6 switches chained together, each switch connecting to 6-7 DUTs (each DUT can have more than one copper port so thats why there aren't many DUTs per switch). The switch at the top of the chain has the only link to external network and all the other switches in the chain use this so that DUTs can access the external network.

Is there any other issue with this network topology especially if more DUTs will be added in the future? One issue I can find at the moment is if one of the switches in the middle of the chain goes down, all other DUTs from that point downwards to the chain will be inaccessible. I am trying to justify why a different topology would be required for this network so that new equipment can be purchased.

Answer 1

Currently we have 6 switches chained together,

Don't do that. Long chains, rings, double rings in switching require special (often proprietary) protocols instead of spanning-tree. Some of these sail in product ranges under the "industrial Ethernet" flag.

The two main drawbacks of a chain, you already listed:

• the failure domain is large. One switch along the chain going down will split the network
• if you don't run a protocol like one of the spanning-tree varieties (or a proprietary cousin thereof), you have no protection against a loop, and with that length of the chain, you're already nearing what STP can handle properly.

Here's another:

• in a chain, overall network capacity remains limited to the most loaded link's capacity

I am trying to justify why a different topology would be required for this network

Generally, a star/tree (or cascaded tree) topology has many advantages:

• placing the root bridge comes naturally
• the question where to attach an external link finds a "natural" answer: near/at the root.
• a switch's backplane performance is usually well above any single link's performance; overall network capacity is boosted to (at least) the sum of all link capacities attached to the root bridge.
• overall network capacity/bandwidth can be evenly spread out; each link into a branch/subtree only needs to carry that branch's traffic/load (and not everyone else's, too).
• branches of the tree can be added or removed easily with little worry and near-to-no impact on the rest of the network
• failure domains get smaller (limited to one branch or sub-branch, usually)
• [although not really recommended] A star/tree/cascaded tree can be built using small-scale - even unmanaged - switches, without any of the more advanced protocols, and it can still scale up or out very well
• high availability features (if so needed) can be added more easily by having a switch pair instead of a single switch at the star's center/the root [1]

[1] there's many ways to get two devices to act as a pair, and some advanced protocols will get into the game: vPC, Stack, MLAG, and of course spanning-tree varieties, too.

Answer 2

An addition to Marc's totally valid answer:

I am trying to justify why a different topology would be required for this network so that new equipment can be purchased.

If there are no cabling/geographical restrictions there's no need for additional ports or switches.

Breaking the inadequate chain, selecting one switch as 'core' and connecting all the others (as 'access') to it in a tree uses the exact same amount of ports. The only difference is that the core switches uses more interlink ports and the others less.

Except for neatness, there's no real need to dedicate a whole switch as 'core'. I'd recommend using a clean cut though for 'core' and 'access' ports, e.g. 1-16 as core (with room for expansion), 17-48 as access.

A star or its extended form, a tree is the 'natural' form for an Ethernet network. Reconnecting your switches in that fashion not only makes your network faster, more resilient and easier to manage (see Marc's answer), it also opens up the opportunity to easily expand the network by connecting more 'access' switches.

With the current chain of seven switches you've reached the design limit for STP. Chaining additional switches will break STP and cause a permanently reconverging spanning tree without reliable operation. Tweaking the configuration would allow you to increase the limits but only so much.

Answer 3

To justify anything you need goals in mind.

• What are these "DUTs"? (I'm guessing DUT stands for "device under test" but that still leaves a load of possibilities open).
• What are their reliability requirements?
• What are their bandwidth requirements?
• Is a partially functioning system useful? or will any failure trigger the immediate need for repair?

The main downsides of chaining switches are.

• When a failure happens, there is a much greater chance that it takes out a large chunk of the network.
• Ethernet has a protcol called "spanning tree protocol" which can be used to provide redundany in the event of link failures, this works best when the topology change caused by the failure is small and it is also places a limit on the length of chains.
• Latency can be higher, but this is unlikely to be a practical issue in all by the most sensitive of environments.
• If high bandwidth is required then all of the links in the chain need to be high bandwidth, whereas a more star like topology may only need a relative handful of links to be high bandwidth.

The question you have to ask is what if-any of those downsides are actually relevant to your application.