Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: [EFM] Re: OAM - To side-band or not to side-band



Roy,

Perhaps you'd better get your 802 terminology straightened out before we continue this discussion - see my comments interspersed below. Reading the latest rev of the 802 Overview & Architecture would seem like a terribly useful first step - perhaps also reading the 802.1D MAC Bridge standard would also be of help to you in understanding what a MAC Bridge is, and what a switch is not. You will find the latest rev of the 802 Overview & Architecture at:

http://www.ieee802.org/1/pages/802.html

Username: p8021
Password: go_wildcats

802.1D is freely available via the "Get 802" scheme - go look at the IEEE "Standards Online" webpage:

http://standards.ieee.org/getieee802/

Regards,
Tony

At 17:37 26/09/2001 -0500, Roy Bynum wrote:

Tony,

Again, I think you may not be understanding what I am talking about.

Actually, according to Geoff Thompson, there is no such thing as a "full duplex repeater".

I believe that is true in the context of what 802.3 has defined to date. The 802 architecture certainly doesn't preclude such a device, if there is a need for one to be defined (see below).

  A hub is not a repeater, it is a half duplex bridge.

Not true, and I very much doubt that Geoff would have said otherwise.  A hub (at least in the context of 802) is nothing more than a multi-port repeater.

To quote from the latest revision of the 802 Overview and Architecture:

"Standard methods of interworking fall into three general categories, depending upon the layer at which the
corresponding interconnection devices operate:
— Physical-layer interconnection, using devices usually termed
repeaters or hubs (6.3.1);
— MAC-sublayer interconnection, using devices termed
bridges (6.3.2);
— Network-layer interconnection, using devices usually termed
routers (6.3.3)."

And quoting further from 6.3.1:

"A repeater is a device used to interconnect segments of the physical communications media, for example to extend the range of a LAN when the physical specifications of the technology would otherwise be exceeded, while still providing a single access domain for the attached LAN stations. Repeaters used in support of multiple end stations attached by star-wired network topologies are frequently referred to as hubs."

So, according to 802 usage, a hub *is* a repeater, and *is not* a Bridge. Hubs can be (and were/still are) built with no link layer functionality whatsoever, although managed 802.3 hubs (as I pointed out) do also exist, managed using  MAC frames. As Andrew Smith points out in his response, you should not make the mistake of confusing the abstract objects that we write standards about with what goes into actual product specifications; if I choose to put an 802.3 MAC in the same box as an 802.3 hub in order to allow it to be managed, that is up to me as a product developer.

An 802.3 managed hub is therefore logically nothing more than a repeater that happens to be packaged with a  management end station in the same box, the end station being connected to one of the hub's ports. The repeater functionality of such a product is still just a layer 1 function.

  A switch is a full duplex "bridge".  Both have 802.1 functionality.

Actually, "switch" is not a defined term in any 802 standard that I am aware of; it is certainly not defined in the 802.1D Bridge standard. It is a colloquial term that, to many, means "A Bridge, having only full duplex Ports"; however, I have also seen/heard it used to refer to hubs, and also to multi-port Bridges that have half-duplex ports. 802.1 makes no distinction between Bridges that have half duplex Ports, Bridges that have full duplex Ports, and Bridges that have a mixture of the two. Whether the MACs are full or half duplex is of no concern to the Bridging function. Again, to quote from the latest revision of the 802 Overview and Architecture:

"The term switch is often used to refer to some classes of bridge. However, there is no consistent meaning applied to the distinction between the terms bridge and switch, and ISO/IEC 15802-3 does not make any such distinction. Hence, this standard only uses the term bridge."

(ISO/IEC 15802-3 is, of course, just the ISO version of IEEE 802.1D - the MAC Bridge standard.)

What I am referring does not have any 802.1 or 802.2 functionality. 

If you decide to outlaw the use of link layer functionality in a product that performs a repeater function, then the only way to manage it would be to use a PHY-based communications channel; however, as I point out above, this is an implementation choice.

Actually, while we are being picky about definitions, the OSI reference model, or specifically, the OSI Management Framework (part 4 of the OSI RM) defines management to be an application layer activity; it is therefore arguable that what you want to do (manage a device that has no functionality above layer 1) is itself a contradiction in terms, as the fact that you are performing management means that, by definition, you are using all 7 layers of the OSI RM (although some of the intervening layers might have only rudimentary functionality, as necessary to support the management application). Putting it another way, if you are talking about a device that has no functionlity above layer 1, then that device, by definition, is not, and cannot be, manageable, because if it is, the device will have layer >1 functionality of some kind.

It has no visibility into the revenue data traffic stream.  It can not get access to the revenue traffic data stream to put upper level application management traffic such as SNMP into the revenue traffic data stream in either direction.

Thank you,
Roy Bynum



At 07:47 PM 9/26/01 +0100, Tony Jeffree wrote:
Roy -

Managed Ethernet repeaters (more commonly known as hubs these days) have been around for some while, and they use MAC frames to carry their (SNMP) management exchanges. I would therefore hesitate to use that particular argument either for or against the use of a side-band for OAM.

Having said that, in networks with repeaters there may be distinct advantages in *not* using such a side-band for OAM - for example, where it is the device the other side of the repeater that you want to manage. Unless, of course, you start putting some form of addressing into this PHY-based side channel, which rapidly starts to look like you're replicating MAC functionality in the PHY, which begins to look like a waste of time & effort.

As to T1 and T3, there's no doubt that some of the EFM participants (myself for one) are not intimately acquainted with the management entrails of these technologies, and with the thinking behind why they are that way. I'm sure that some of that information may be useful in informing what we do in EFM.  However, I'm equally sure that re-inventing T1 or T3, giving it a bit-rate that is a power of 10, & then badging it "Ethernet", would be a total waste of our time.  After all, if you continue to do what you have always done, you inevitably end up with what you have always had.

Regards,
Tony


At 11:31 26/09/2001 -0500, Roy Bynum wrote:
There are a lot of other reasons to have the OAM ou-of-band to the MAC traffic, such as being able to support OAM on an intelligent "transparent" full duplex repeater in the future.  When this group took on the task of adding subscription network support for edge access infrastructure into Ethernet, they took on applying most all of the functionality that is being used today.  There is a long history of why the functionality for these types of services is what it is.  How many of the EFM Task Force people have looked at how the OAM overhead of T1 or T3 framing works today?  How many of the EFM Task Force people have looked into why the OAM overhead of T1 or T3 framing works the way that it does?