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

RE: [EFM] OAM - Faye's seven points




Roy -

I have a great deal of sympathy with Andrew's questions.

Apparently, now is the time to agree that a "side band" OAM channel should 
be a fundamental part of the EFM standard, but now is not the time to look 
in detail at the requirements that might lead us to choose that solution 
over the other alternatives, including a frame-based OAM mechanism using 
the existing MAC service. This seems to me to be a bad case of 
inappropriate ordering of carts and horses - in the design processes that I 
have been used to hitherto, the starting point was not to define the 
solution & then retrofit the justification, but to start by identifying the 
requirements that will drive out the characteristics of the solution.

As far as I can tell from the thread so far, the "requirements" that have 
been identified for this side-band are no more than a statement that it 
should utilize some arbitrary but fixed (in a given application) proportion 
of the bandwidth available on the medium concerned, and that it should 
display deterministic properties with respect to its transmission 
characteristics (i.e., it should be a "synchronous" channel rather than 
"asynchronous"). However, I have yet to see:

- A clear statement of what proportion of the bandwidth will be required 
for OAM, or whether it is a proportion of the available bandwidth or just a 
fixed overhead (e.g., X Kbit/s, regardless of the size of the pipe), or 
even a clear statement that it will not be possible ever to make any hard 
and fast statements about this, and why;

- A clear statement of why there is a need for deterministic 
characteristics in the OAM "channel", other than oblique hints to as yet 
unexplored and unspecified aspects of performance management.

I suspect that the short answer to this first question is that there are no 
hard-and-fast answers; therefore, if the side channel approach is adopted, 
it will presumably either be based on a finger-in-the-air choice, which 
will later prove to be broken or over-provisioned depending on the 
application, or will have to be configurable to suit the application. (I 
will leave it to others to comment on whether offering an OAM side band at 
the PHY level, and making the available bandwidth of the side band 
configurable, is desirable/feasible/practical/economically viable. To 
paraphrase Bob Donnan, who once famously commented that he gets nose-bleeds 
when he ventures above the data link layer, I get the bends if I spend too 
much time down in the PHY).

The second question seems to me to be particularly pertinent, as, from what 
I have gleaned from your response to my "free lunches" Email, this seems to 
have the potential to be a pivotal reason for choosing of a side band 
rather than simply using MAC frames (MAC control or otherwise) for OAM 
traffic.  It also reminds me very much of the religious wars of the early 
'80s between the supporters of 802.4 on the one hand, arguing that 
determinism was an essential characteristic of LAN communication, 
particularly for the requirements of process control and other critical 
"real time" applications, and the supporters of 802.3 on the other hand, 
arguing that this was not so, and that all you needed was 
Ethernet.  Suffice it to say that today, 802.4 is a little-known blind 
alley in the progress of LAN technology, while 802.3/Ethernet networks are 
now widely deployed in all aspects of industry and commerce, including 
process control (and other critical real time) applications.

For me, arguments of the form that we've got to do it this way be able 
"...to support a wide diversity of applications" really don't cut it, 
unless they can be backed up. And I don't believe that it is necessary to 
enumerate all the possible OAM applications that might ever be imagined in 
order to reach a conclusion here, even if it were possible to do such a 
thing; it is simply necessary to identify the "killer app(s)", if there are 
any, that because of their characteristics, mean that we have to go a 
particular way. If there are no such killer apps, then I would question the 
need to spend any further cycles on the "side band" discussion, and would 
suggest that we move on to more fruitful discussion areas.

Right now, I am personally left with the strong impression that the major 
reason why a side band is being proposed is that this is what the Telcos 
expect, and that, given they are a significant part of the target market, 
there is a marketing obstacle (rather than a technical obstacle) in the way 
of doing it by some other means. If I am wrong, please convince me otherwise.

Regards,
Tony



At 19:35 24/09/2001 -0500, Roy Bynum wrote:

>Andrew,
>
>There is a time for everything.  Right now is not the time.  There are 
>other, more fundamental issues that need to be addressed.  As I have time, 
>I will be doing presentations on various OAM functions that need to be 
>covered.  There also needs to be an opportunity for others to use their 
>imaginations.
>
>Thank you,
>Roy Bynum
>
>
>At 05:48 PM 9/24/01 -0700, Andrew Smith wrote:
>
>>Roy,
>>
>>I think you misunderstand my intent. You know far more than I do about the
>>services that you want to deploy and their OAM requirements. I am trying to
>>goad you into sharing your experience but you do not seem to want to do so.
>>I guess that's the end of this thread then.
>>
>>Respectfully,
>>
>>Andrew Smith
>>
>>-----Original Message-----
>>From: Roy Bynum [mailto:rabynum@xxxxxxxxxxxxxx]
>>Sent: Monday, September 24, 2001 5:13 PM
>>To: ah_smith@xxxxxxxxxxx
>>Cc: stds-802-3-efm
>>Subject: RE: [EFM] OAM - Faye's seven points
>>
>>
>>Andrew,
>>
>>By your attitude, I take it that you do not want be able to support a wide
>>diversity of applications.  I suppose that you want to limit the service
>>deployment of EFM and the market that it might achieve.  I guess that you
>>want limited Ethernet services that can only support IP services and do not
>>want to see an expansion of the service models that can be achieved with
>>Ethernet.   In stead of trying to goad me, why don't you start to enumerate
>>all of the services, from "Private Line" to interactive Video calling that
>>you can think of, and define the service/operations overhead that each
>>would need.  Perhaps then, something productive will come of this thread.
>>
>>Thank you,
>>Roy Bynum
>>
>>At 03:47 PM 9/24/01 -0700, Andrew Smith wrote:
>> >Roy,
>> >
>> >Well, if you won't take the time to quantify the requirement, I think
>>you'll
>> >have a hard time convincing this committee to adopt what you propose.
>> >
>> >Andrew Smith
>> >
>> >
>> >-----Original Message-----
>> >From: Roy Bynum [mailto:rabynum@xxxxxxxxxxxxxx]
>> >Sent: Monday, September 24, 2001 2:19 PM
>> >To: ah_smith@xxxxxxxxxxx
>> >Cc: stds-802-3-efm
>> >Subject: RE: [EFM] OAM - Faye's seven points
>> >
>> >
>> >Andrew,
>> >
>> >What amount of OAM bandwidth per user would a copper local loop at 2mbps
>> >revenue bandwidth rate need?  How would having different services models
>> >change that OAM bandwidth?  These are questions that can be "nit-picked" to
>> >death.  This is very early in this process to trying to get to this level
>> >of detail for each and every potential type of service and deployment
>> >architecture model that can be imagined.  I can think of quite a few
>> >variations myself, that I will not take time to enumerate here,  one, I
>> >don't have the time, secondly, because I think the different vendors will
>> >want to do that themselves.  I am attempting to find a reasonable answer
>> >that will cover as many of the variations of services and topologies as
>> >possible.  I am sorry that you do not seem to agree with that approach.
>> >
>> >Thank you,
>> >Roy Bynum
>> >
>> >At 01:57 PM 9/24/01 -0700, Andrew Smith wrote:
>> > >Roy,
>> > >
>> > >I'm still not clear whether you are arguing that the OAM rate should be
>> > >proportional to the line rate (I think you said 0.1% in your slides) or
>> > >whether you propose an absolute value(you write 1 Mbps below). I haven't
>> > >seen any argument in support of the OAM needs increasing in proportion to
>> > >the line rate. Terms like "very low" and "reasonable" are hard to judge
>>and
>> > >"tends to provide for a wide diversity of services and infrastructure
>> > >topologies" sounds very nebulous: how about some hard data?
>> > >
>> > >Andrew Smith
>> > >
>> > >-----Original Message-----
>> > >From: Roy Bynum [mailto:rabynum@xxxxxxxxxxxxxx]
>> > >Sent: Monday, September 24, 2001 12:36 PM
>> > >To: ah_smith@xxxxxxxxxxx
>> > >Cc: stds-802-3-efm
>> > >Subject: RE: [EFM] OAM - Faye's seven points
>> > >
>> > >
>> > >Andrew,
>> > >
>> > >OAM bandwidth can be very low for simple, single treaded services such as
>> > >Internet access.  For diverse multiple services the OAM bandwidth needs
>>to
>> > >be relatively high.  If history is any guide, whatever we do, it will not
>> > >be enough long term.  At present, what we should be looking for is a
>> > >reasonable balance of OAM overhead to revenue bandwidth.  In my
>> > >presentation http://www.ieee802.org/3/efm/public/sep01/bynum_2_0901.pdf,
>>I
>> > >suggested an overhead bandwidth of about 1 Mbps for a revenue bandwidth
>>of
>> > >1Gbps.  I believe that to be a reasonable amount of OAM bandwidth.  It
>> > >tends to provide for a wide diversity of services and infrastructure
>> > >topologies.  Does it need to more than that?
>> > >
>> > >Thank you,
>> > >Roy Bynum
>> > >
>> > >At 06:59 PM 9/21/01 -0700, Andrew Smith wrote:
>> > >
>> > > >Roy,
>> > > >
>> > > >Well, you can't choose "all of the above": what do you think is the
>>right
>> > > >interval for your OAM needs and why? It wasn't meant as a rhetorical
>> > > >question.
>> > > >
>> > > >Andrew Smith
>> > > >
>> > > >
>> > > >-----Original Message-----
>> > > >From: Roy Bynum [mailto:rabynum@xxxxxxxxxxxxxx]
>> > > >Sent: Friday, September 21, 2001 6:18 PM
>> > > >To: ah_smith@xxxxxxxxxxx
>> > > >Cc: stds-802-3-efm
>> > > >Subject: RE: [EFM] OAM - Faye's seven points
>> > > >
>> > > >
>> > > >Andrew,
>> > > >
>> > > >Yes to all of the below.  If by a "scheduler" you are referring to
>>"every
>> >x
>> > > >'revenue whatever' an 'OAM whatever' is inserted regardless of whatever
>> >is
>> > > >happening and without effecting whatever is happening in the 'revenue
>> > > >whatever' and the 'OAM whatever' is not effected by the 'revenue
>> > >whatever'".
>> > > >
>> > > >Thank you,
>> > > >Roy Bynum
>> > > >
>> > > >At 06:31 PM 9/21/01 -0700, Andrew Smith wrote:
>> > > > >Roy,
>> > > > >
>> > > > >You're not being clear: when you say "constant", over what interval
>>do
>> > >you
>> > > > >measure: one Frame? one Byte? one Bit? the time it takes for an
>> >electron
>> > >to
>> > > > >jump from one atomic orbit to another? It doesn't matter which one of
>> > >these
>> > > > >you choose, you still need a *scheduler* to put the bits of your
>> >message
>> > >(I
>> > > > >assume that your OAM messages are more than one bit long) onto the
>> > >medium.
>> > > > >
>> > > > >Andrew Smith
>> > > > >
>> > > > >P.S. Please let me buy you lunch someday.
>> > > > >
>> > > > >-----Original Message-----
>> > > > >From: Roy Bynum [mailto:rabynum@xxxxxxxxxxxxxx]
>> > > > >Sent: Friday, September 21, 2001 5:46 PM
>> > > > >To: ah_smith@xxxxxxxxxxx
>> > > > >Cc: stds-802-3-efm
>> > > > >Subject: RE: [EFM] OAM - Faye's seven points
>> > > > >
>> > > > >
>> > > > >Andrew,
>> > > > >
>> > > > >A "side band" is a constant bandwidth facility.  Most of what you are
>> > > > >referring to only would apply to an in-band, "frame" based OAM.  It
>> >does
>> > > > >not apply to a "side band" OAM channel.
>> > > > >
>> > > > >Thank you,
>> > > > >Roy Bynum
>> > > > >
>> > > > >At 05:53 PM 9/21/01 -0700, Andrew Smith wrote:
>> > > > > >Roy,
>> > > > > >
>> > > > > >I think we're talking past each other here (see Tony's lunchtime
>> > > >comment).
>> > > > > >
>> > > > > >Implementation of a "side-band" channel *requires* a scheduler and
>> > > >queueing
>> > > > > >of its own. The side-band method is the one that adds the unneeded
>> > > > > >complexity by mandating an additional scheduler on top of the ones
>> >used
>> > > >by
>> > > > > >higher layers that (in any reasonably designed piece of EFM gear)
>> >will
>> > > > > >already be present.
>> > > > > >
>> > > > > >I challenge this group to come up with appropriate dimensions for
>> >such
>> > >a
>> > > > > >side-band channel - what peak or sustained bandwidth? what burst
>> > >size? -
>> > > > > >that does not cause EFM to become an evolutionary dead-end.
>> > > > > >
>> > > > > >Andrew Smith
>> > > > > >
>> > > > > >
>> > > > > >
>> > > > > >-----Original Message-----
>> > > > > >From: Roy Bynum [mailto:rabynum@xxxxxxxxxxxxxx]
>> > > > > >Sent: Friday, September 21, 2001 3:53 PM
>> > > > > >To: ah_smith@xxxxxxxxxxx; Tony Jeffree
>> > > > > >Cc: stds-802-3-efm
>> > > > > >Subject: RE: [EFM] OAM - Faye's seven points
>> > > > > >
>> > > > > >
>> > > > > >Andrew,
>> > > > > >
>> > > > > >What you are referring to in the need for "sort of token bucket
>> > > > > >scheduler",  and "...want to allow the OAM "channel" an unfair
>> > >advantage
>> > > >in
>> > > > > >the use of spare bandwidth too, implying some sort of priority in
>>the
>> > > > > >scheduler" would only apply if the OAM were "frame" based.  If the
>> >OAM
>> > > >were
>> > > > > >a "side band", or "out-of-band" to the revenue traffic, the all of
>> >that
>> > > > > >complexity is unneeded.
>> > > > > >
>> > > > > >With an OAM "out-of-band" channel, the OAM bandwidth is
>>predetermined
>> > >by
>> > > > > >the bandwidth of the "side band" data.  It would also not interfere
>> > >with
>> > > > > >the revenue bandwidth.
>> > > > > >
>> > > > > >Thank you,
>> > > > > >Roy Bynum
>