Which access methods and topologies are used for 10/40/100 Giga Ethernet?

Question

So I'm learning about different Ethernet types and it's stated that the Standard Ethernet operates at Bus/Star and uses CSMA/CD as an access method. The same is stated for Fast Ethernet and Giga Ethernet minus the Bus part. I was wondering about 10/40/100 Giga Ethernets? I have looked everywhere but couldn't find a single clue except "All Gigabit Ethernet variants use a star topology" regarding the topologies. Is it safe to assume that these are considered Gigabit variants? and what about the access methods? I can't imagine they're using CSMA/CD because it sounds like they've started moving towards full-duplex mode at Giga Ethernet

Answer 1

CSMA/CD is obsolete. It was only used with 10 and 100 Mbit/s Ethernet. CSMA/CD was specified for Gigabit Ethernet as well but wasn't used anywhere. All modern variants use full-duplex, point-to-point links without requiring a channel access method or creating collision domains.

Also, all modern variants (except for PON variants) use point-to-point links, requiring switches for connecting more than two nodes. A single switch arranges the attached nodes in a star topology and multiple switches are connected in a tree topology. Chains or rings are also possible using a spanning tree protocol, but less optimal.

Answer 2

A medium access control method is only needed if there is the possibility for conflict between multiple simultaneous transmitters.

Over time the cost of switches that were able to buffer frames came down and they gradually replaced the dumb repeater hubs that were initially used to build twisted pair Ethernet networks. The combination of switches that could buffer frames and full duplex physical layers meant that Medium access control was no longer needed.

• 10 megabit Ethernet networks were nearly always half-duplex with CSMA/CD, either with coaxial cables or with dumb repeater hubs.
• 100 megabit Ethernet networks were a mixture. Early 100 megabit networks often used dumb repeater hubs and CSMA/CD, but by the end of the 100 megabit era switched full duplex networks had become the norm.
• Gigabit Ethernet supported CSMA/CD in theory, but I've never seen any evidence that it was actually used in practice.
• The 10 gigabit and newer Ethernet standards dropped CSMA/CD support completely, all links are full duplex and there is no medium access control.

---

Ethernet bridging is designed around a loop-free topology (a star or tree). A loop free topology makes it easy to avoid duplicating packets and means that ingress ports can be reliably used to track the location of a host.

However loop free topologies have reliability and scalability problems. A variety of techniques have been used to mitigate these.

• Spanning tree protocol allows the physical network to have loops, but disables links so that the active part of the network is loop free. This mitigates some of the reliability issues of loop-free networks, but it doesn't help with the scalability issues.
• Link aggregation allows multiple physical links between the same two devices to be combined into a single logical link. This helps with some reliability and scalability issues, but still leaves the network vulnerable to switch failure.
• Most major vendors offer some form of "multi-chassis link aggregation". Two (or potentially more) switches of the same brand can be linked together and act as a single logical switch for link aggregation purposes. This provides some defence against switch failure without leaving links sitting unused.

Even with all these mitigations though, as networks grow beyond a certain size/traffic volume, the loop-free topology requirements of Ethernet bridging become more and more of a limiting factor.

IP routing on the other hand has absolutely no need for a loop-free topology. Thus you will often see networks where Ethernet links are simply used to make point to point connections between IP routers (or "L3 switches"), totally sidestepping the topology limitations of Ethernet.

In some cases you will even see situations where a virtual Ethernet network is built on top of an IP network, which in turn is built on top of point to point Ethernet links.