4x 10Gbps Port Total Bandwidth Questions

[Ksec]

Apology if this is a stupid question.

Is it theoretically possible to have a 4 Port 10Gbps Port Switch, where each all 4 cable would work and connect with 10Gbps identify / shown, but the total switching speed is limited to 20Gbps?

So if all port were trying to max out the transfer, they would only get 5Gbps each. And if only 2 where trying to max out they will get 10Gbps each.

Is this within the spec and realm of possibilities?

Edit: Adding more context.

I was thinking more in the Cost of 10Gbps Ethernet. Or It could be Nbase-T 5Gbps Ethernet. For consumers, 95% of our time would not need anything beyond 1Gbps. But for the rare occasion of NAS File transfer or If you have 1Gbps Internet . That would be very useful. It wouldn't make sense to have a Switch / Router to support 4x 10Gbps Ethernet in full, but if As long as it allows in rare occasion to go with 10Gbps that would be enough for most of that usage.

So I was wondering if something like that could theoretically exist and save cost. 4x 10Gbps Port but it could only run at Max 15Gbps. or something similar. The question really was born out of consumer equipment got stuck with 1Gbps. And anything above is ridiculously expensive.

 

[RTM]

There is no such thing as a stupid question, though yours may be a little weird
A lot of stuff is theoretically possible, and yes I see no reason why a switch like that could not exist.

Perhaps you want to explain why you are asking, it will help us give meaningful answers.

Are you looking for a solution, where you for some reason limit overall traffic to 20G
in which case this could probably done via QoS traffic shaping

Are you troubleshooting an issue where you appear to have issues with bandwidth?
In which case, I would argue that it is possible, but assuming we are talking L2 switching and at least a fairly recent switch it is unlikely.
It could be QoS implemented somewhere

 

[Scott Laird]

There are a few weird assumptions in your question; there's no real reason that 4x links all talking at full speed need to cause a problem. Modern Ethernet is full duplex, so pairs of systems could talk to each other at full speed through the switch without congestion. Imagine you have systems A, B, C, and D connected to the switch, and A sends 10 Gbps to B, B sends 10 Gbps to A, C sends 10 Gbps to D, and D sends 10 Gbps to C. That shouldn't be a problem at all. There's nothing magical about this pattern, it's just the easiest example. In these cases, you could move 40 Gbps total through the switch.

On the other hand, imagine that system A sends 10 Gbps of broadcast or multicast traffic; that'd eat 100% of the outbound bandwidth from the switch on the other ports. Or, in the opposite direction, systems B, C, and D all want to talk exclusively to system A. In either case, you're going to be limited to 10 Gbps total.

There's nothing special about 20 Gbps total, or 5 Gbps per port. It all depends on traffic patterns. Given that, no one is going to build a switch that inherently limits bandwidth out of the box, although higher-end switches may be able to apply traffic shaping to artificially induce this sort of behavior. I've never done L2 traffic shaping, though--that's usually more of an L3 thing, and I'm not sure what hardware/software would be able to do L2 shaping. You'd probably be better off doing it on the hosts, if it actually matters for some reason.

Practically speaking, I can't see how it'd matter in a general-purpose system though. It's not like any of the systems would starve if you overload things. If you try to send more than 10 Gbps of traffic to any port, then the switch will either drop traffic more or less randomly or trigger Ethernet flow control on the inbound links that are flooding the switch. In either case, software should be able to cope. TCP will back off and slow down, and everything should balance out just a bit shy of full utilization. So, your sign of overload will simply be that things get a bit slower but still work just fine. If you really *need* 10 Gbps, then this would be your signal to move to faster links and/or a bigger switch.

The only case where I could see this sort of behavior actually being a problem would be for something like RoCE, which really hates having any Ethernet frames lost. From what I've seen (admittedly not a lot), keeping RoCE happy (especially in non-trivial networks) kinda counts as an Advanced Networking Skill.

 

[Ksec]

There is no such thing as a stupid question, though yours may be a little weird
A lot of stuff is theoretically possible, and yes I see no reason why a switch like that could not exist.

Perhaps you want to explain why you are asking, it will help us give meaningful answers.

Are you looking for a solution, where you for some reason limit overall traffic to 20G
in which case this could probably done via QoS traffic shaping

Are you troubleshooting an issue where you appear to have issues with bandwidth?
In which case, I would argue that it is possible, but assuming we are talking L2 switching and at least a fairly recent switch it is unlikely.
It could be QoS implemented somewhere

I have added more context to the question. Thank You.

 

[PigLover]

The scenario you pose is not at all unusual. It is quite common to have systems that are "oversubscribed" (i.e., have more raw IO capacity than can be processed by the system). In fact, many switches on the market today that advertise themselves as being "non-blocking" on throughput are actually over-subscribed when looking at processing that throughput in the form of of small packets. Take a look on Mikrotik's web site where they publish performance data on many of their switches and you can see this quite clearly.

That said - you really see this most often on larger switches. At the scale you are talking about (4 ports) there is little/no advantage. Silicon switch fabric that can run at 40-80Gbps is not costly - most of the cost of the switch at that scale is in the PHY (the physical IO interface at each port).

So the concept of your question is valid - but at the scale you propose it is an academic exercise rather than something likely to have practical application.