5

On a packet switching network, a router is required to route packets to different networks.

Why does packet switching uses the term 'switching' when a packet is a term for a layer 3 PDU and switching is layer 2?

3

Years ago, in a Cisco course (BCMSN, when it existed) the teacher explained to us the Evolution of the LAN, from the ThickNet (Coax) to the current Switched UTP/Fiber connection

At first, various LAN were either routed connected with Hubs, with a hub being often prefered because of cost if possible. When a router wasn't really needed but a Hub didn't do the job, they split the LANs with Bridges.

HOST(A)----\
HOST(B)----(HUB)-------HOST(D)
HOST(C)----/

When the bridge saw a frame going by, it noticed it was from SRC port X to somewhere and noted it. Since it didn't know where the DST was it broadcasted on the other port (early bridge had 2 ports it was simply used to limit broadcast domain) When it got the reply back, it noted from where it was coming from etc. Note that you had broadcast medium on each side (Coax)

NETWORK(A)---(BRIDGE)---NETWORK(B)

cost less than

NETWORK(A)==(Router)==NETWORK(B)

but not useful everywhere

Then, someone had the idea to bridge EACH station to a central network... all in the same box, with the network in the middle to pass data between interfaces. This also means that it's easier to track hosts to a specific port, thus limiting flooding to actual Unknown MACs or Broadcast MAC. It also limited the collision domain, as no two hosts were directly talking to each other, Full Duplex also became possible.

HOST(A)---\----------------------/---Host(D)
          | Bridge        Bridge |
          |    \            /    |
          |     ------------     |
          |     | INTERNAL |     |
HOST(B)---|Br.--|  SHARED  |--Br.|---HOST(E)
          |     | NETWORK  |     |
          |     ------------     |
          |     /          \     |
          | Bridge        Bridge |
HOST(C)---/----------------------\---ROUTER(A)

One last problem was that you had limited bandwidth in the internal network as it created a single bottleneck for all traffic in the case where it couldn't transfer data from one port to another quickly enough (Buffers were used, but they had their limits). To remediate to that, they devised special hardware to optimize inter-port traffic and instead of using an internal shared network, have the equivalent of a Phone system's "Switchboard" where you can temporarily connect two ports without blocking data connection between another set of two ports

Ex: Where Host(A) and Host(D) can communicate full speed while Host(C) communicates with Router(A) and one doesn't block the other nor reduce their possible bandwidth

This was called a "Switched Fabric", and thus was born The Network Switch

            --------------
HOST(A)-----|  INTERNAL  |-----SWITCH(B)
            |            |
HOST(B)-----|   SWITCH   |-----HOST(D)
            |            |
HOST(C)-----|   FABRIC   |-----ROUTER(A)
            --------------

So, that is why we say a frame is Switched - the temporary (virtual) interconnection of two ports being like the old Telephone Switchboard, for the duration of the frame transfer. (It's all done by ASICs with buffers and very fast memory now, but you get the gist of it)

Hope it helped!

| improve this answer | |
  • I do not get it starting from the forth diagram. I hope you can update the diagram. – Ron Vince Sep 26 '14 at 12:40
  • Clarified things a bit... (diagram & wording) – Remi Letourneau Sep 26 '14 at 13:51
  • If I understand the 4th diagram correctly, the connection between bridges (internal shared network) is similar to Ethernet on coaxial. Is it correct? – Ron Vince Sep 28 '14 at 2:58
  • That was the first concepts, at least, it's how the teacher told the story. – Remi Letourneau Sep 29 '14 at 12:33
7

A very good question coming at it from a networking viewpoint. Your question is based on the networking models that were developed long after packet switching was proposed as a concept. However networks across any distance were interconnected over the existing telecommunications "networks" and while many network and telecommunications terms are the same, they will carry somewhat different meanings.

Traditional phone connections are circuit switched, meaning that there is one established circuit forming a connection between two end points. A telephone switch in a phone system used to be the device that created and tore down these connections. If you go way back, human operators would work at "switch boards" manually making and removing these connections.

Packet switching is a much more recent concept (relatively speaking) and refers to networks that would take the data and transmit the data units independently over the network. It is often far more efficient to "share" the resources of the underlying network in this way. Today, VoIP uses "virtual circuits" to emulate circuit switched connections over a packet switched network.

The Wikipedia pages for Circuit Switching and Packet Switching are pretty well written and should provide more background.

| improve this answer | |
  • In circuit switching, there is no fragmentation of data, right? – Ron Vince Sep 23 '14 at 6:04
  • That is not necessarily true. You can have a circuit switched connectionn (say ISDN) over which you are sending IP traffic. This could most certainly be fragmented as per normal networking rules. The terms packet switching and circuit switching have more to do with the type of connection of the underlying network, and less about the actual networking itself. – YLearn Sep 23 '14 at 14:14
  • In pure circuit switching, is there fragmentation of data? Because I read that there is no layer model. – Ron Vince Sep 24 '14 at 1:12
  • This is true of the underlying circuit as the circuit is a guaranteed connection from one end to the other with a fixed amount of bandwidth. As such, there is no need for fragmentation within the circuit switched network. – YLearn Sep 24 '14 at 2:06
1

A packet (L3) still needs to be switched as a frame (L2) as it ingresses and egresses interfaces. A router would serve no useful purpose if it makes routing decisions, but has no way to switch packets out. In other words, you can take routing out of switching, but you can't take switching out of routing.

| improve this answer | |
  • 1
    Not really sure how this applies to packet switching, but it isn't really accurate either. Routers existed long before "switches" and worked just fine. A router to router connection has L2, but there is no switching involved. Even today, you could connect your router to a hub, again avoiding switching. – YLearn Sep 23 '14 at 14:19

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.