Connectionless mode

Bonjour,

Indeed Michelot.

Broadcast is multipoint-to-multipoint, a process in connectionless networks.

But, it exists oriented connection Ethernet, e.g. PBB-TE.

Best regards, Michelot

Salut, Michelot,

I suppose that PPPoE can also be considered a connection-oriented way of using Ethernet. I think, though, that ideas like PBB-TE are more in the vein of using Ethernet-formatted frames in non-Ethernet environments.

On the other hand, you might say the same thing about switched Ethernet, as opposed to the original CSMA/CD Ethernet. In the switched Ethernet, what is similar to the original shared one, other than the frame format? Well, I guess you could say that the switches are programmed to emulate the behavior of the original shared version, and to give you some extra bonus features too (like learning and IGMP snooping).

"It's complicated."

Bonsoir, Albert

I suppose, unless it is 802.2 Type II traffic.

rick jones

Bonsoir Albert,

Ethernet is connectionless through its broadcast when the destination address is unknown in a bridge forwarding table. Ethernet is connection-oriented when all the bridge forwarding tables on the SA-DA path know the DA address, either by learning, either statically.

It is an hermaphrodite behaviour, male first and female after.

Good question, let's try a reply.

PPPoE is a specific protocol (a single layer). A frame of this protocol is encapsulated in a MAC frame. So, when a MAC frame is sent with a PPPoE frame in its payload, the MAC frame has to find the DA equipment. It can use broadcast to interpellate the destination. If the destination receives the frame, it replies to SA and its address is learned by the bridge. From this point, Ethernet becomes connection- oriented.

I don't know what is non-Ethernet environment. Ethernet 802.3 is made of 2 layers MAC or ETH, then PHY or ETY. These both layers are independant, so ETH could be over other layers, as MPLS, SDH...

Best regards, Michelot

Bonsoir Rick,

Thanks for reminding me that.

I think that the LLC behaviour is independant from the MAC behaviour, it's another layer, as it is for IP over MAC. IP never becomes oriented-connection.

Best regards, Michelot

No - there is no need to set up a connection, and no "state management" needed in the end points.

Bridge filtering is a very successful hack to allow an Ethernet to consist of multiple types of link at different speeds, and to increase the aggregate capacity.

The whole point of transparent bridging design is that it does not change the behaviour needed at the end points.

so all the comms, point to point or otherwise, is connectionless.

Bonjour Stephen,

You're right, there is no need to set up a connection by a signaling protocol. There is an emulation of a connexion because the multipoint- to-multipoint becomes point-to-point (or point-to-multipoint in multicast).

What do call "state management"?

It's perhaps true in LAN but it's different in WAN transport where the user need services.

Have you an example?

What do you make of the traffic conditioning and the traffic shaping ?

Best regards, Michelot

Michelot wrote: (snip)

It is more like a cache that, statistically, makes it likely that the frame only goes to one, or a small number, of destinations.

Endpoints still have to test the destination address, just as they would with a traditional ethernet.

-- glen

last time i checked the crucial difference between connectionless protocols and connection oriented ones is that a "connection oriented" protocol has a state - the device has to handle tracking what is on the other end of each connection.

if you dont keep state information, then you dont have a connection, so you have a connectionless protocol.

now you may try to be clever and do something like MAC bridging that builds caches in the plumbing between the end points - but that is "just" an optimisation technique, and the comms between the end points doesnt care - it works either way.

so the comms protocol is connectionless.

or am i missing something?

True.

And you can build a service layer on top of a connection oriented or connectionless lower layer - it does not imply that the lower layer has to be 1 or the other.

a thought experiment.

2 Ethernet devices communicating - say 1 sends out broadcast packets from time to time and the other can "see" them.

no connection has to be involved.

now add transparent bridging.

the bridges keep state to try to optimise where that MAC traffic needs to go and to save bandwidth.

The bridging function is now keeping cache state - it "knows" which port is on the source path to the MAC address of the broadcast source.

Note there is no information about a connection

The bridge function just has some knowledge about on which port 1 of 2 existing MAC address probably can be reached.

(in a perfect simple world where all simple thought experiments happen all connections are duplex and working and control systems are stable, so source packet reception gives 1 and only 1 path back to that MAC - real life is definitely not perfect).

And if the 2nd MAC does not transmit, the cache function does not detect it, and it may be on any port.....

it only affects volume of traffic, so doesnt change the semantics of what is going on at this level - it just adds more edge cases where things break :)

is just a bunch of engineering compromises between what programmers want to see and what comms engineers can afford to provide......

Bonsoir Stephen,

What is your secret to convince me, normally it=92s difficult!

This is a very good formulation which we can keep in mind.

Now, how do you call the relation between the both end terminations? How do you named the concept similar to a =93network or link connection=94 in a =93connection oriented=94 protocol (or layer).

Personally, I=92m using the expression of =93network or link flow=94 for a =93connectionless=94 protocol, but I=92m not sure that everybody can understand it. This expression is from G.809. We can just used the word of "communication" between the both end points.

I agree with you.

Not missing. I just want to invoke 2 points.

(1) Ethernet 802.3 is not a protocol but a set of 2 protocols. The MAC (or ETH) protocol is connectionless as you have demonstrated it, and the PHY (or ETY) protocol is connection-oriented, as for all physical protocols - both ends have to agree synchronization, modulation, speed, coding=85

(2) It seems there is another crucial difference between connectionless protocols and connection oriented ones. By looking in G.

800 (section 6.5.1), I realised this statement I was not opening up to this idea before.

It=92s about a generic description of a forwarding function (similar to a switch fabric or a matrix inside a bridge).

=93The forwarding function requires control information to identify the output port(s) to which a communication is destined. This control information is carried by the symbol being forwarded (commonly in the form of a destination address). The resulting network behaviour is traditionally known as "connectionless"=94.

In a connected network, these control information are not carried within the frames.

Thanks for your description.

Best regards, Michelot