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

RE: [802.3af] clocking over Ethernet


I agree with you 100%. I know that clocking is in principal out of scope of but please let me just finally clarify the reason for asking in more
detail. I would really appreciate if the discussion could be continued at
another place. Im open for suggestions. Please write directly to me.

Background/Reasons: will be a reason for customers to make changes in their network
infrastructure. These changes over time will be probabbly even major
changes in some places since enables completly new applications. 
Soon we will see cheap solutions and this means: makes it possible to replace industrial busses (profibus/interbus
etc.) by powering sensors/actors, ISDN (by powering phones/gateways) and
propriatary communication/control interfaces.
In my opinion it is even an good alternative (and I know that some consumer
electronic companies work on it) to IEEE1394. I know that 1394 has another
focus but putting a CAT-5 network is somewhat easier than planning a
flexible 1394 network (in Europe we dont have phone nets like in the US).
My house and several of friends have CAT-5 networks at home and I run TV,
audio, phone, alarm system, light/heat control through it.

That ethernet is async is no problem in most applications and I never
intended to change it to isochronous. I just looked for a way to add a sync
signal while keeping COMPATIBLE with existing DTEs. 
The precision I look for is in the sub us or better ns range. This would
enable to sync e.g. mobile phones. It would enable plug and play mobile
(e.g. DECT) networks with support of seamless handover/roaming. Just plug
the basestations into sockets of a synced net. The syncing would improve
radio spectrum use (you wrote that). There are many other applications that
would also benefit from synchronising. Agaoin: the ethernet data transfer
can be kept async. is a reason to develop new switch controllers/PHYs with integrated
support for Since there wont be many further generations of
ethernet this is maybe the last and final chance to introduce sync signals.
A sync enabled DTE could send a round trip signal through two connected
wires like +/. wires: through + to central point and through - back. With
defined transformer parameters it should be possible to measure line length
in order to calculate the delay to have highest possible precision (this
happens once during DTE initialisation). The only thing missing then is a
sync pulse sometimes (once a second is absolute sufficient) from the hub.
The exact clock I can generate in the DTE locally. A sync on the power
avoids any requirement to change ethernet PHYs. A 0.5/0.2 Volt pulse that
is inside the tolerances of would maybe sufficient. A simple
protocol could seperate noise from sync pulses. Im open for other ideas.
But I think there could be a way that can coexist with existing equipment
and still keep the chance for that feature. sync would also enable to add power hubs AND sync for mid span
application without removing eventually expensive switches completly.

Joerg Wilhelm
director R&D 
caporis networks AG * Germany

At 10:05 01.08.01 +0200, Raymond Gass wrote:
>   Bob,     I  remember 802.9. The situation here is a little bit
>different. We are not  necessarily trying to build isochronous services
>over ethernet. Let's take an  example: you connect a DECT or PWT base
>station to your LAN. The base station is remotely  powered according to
>802.3af   Now,  if you don't want to lose timeslots and thus traffic on
>your base stations, they  need to be synchronized. They need a reference
>for the start of  the frame otherwise slots will be lost in the  area
>covered by 2 or more stations (which is needed by essence if you wish  to
>handle  handover properly)   The  first solution is to bring new wires to
>carry sync from a central point to all  base stations. Then,  you lose
>benefit of reusing existing LAN wiring. Not a good  proposal   The  second
>solution: GPS. Nice but problems indoor   Third  proposal, NTP: does not
>work due to transmission time  dispersion   The  proposal from Joerg is
>quite good: you have phantom pairs to carry power, why  not add  SYNC pulse
>on this phantom? Of course, you need  to understand how to cross
>intervening nodes. Those  equipment that do not need power or SYNC are not
>affected (which was  not the  case in .9). There is no change to the
>existing asynchronous nature of Ethernet  as it  currently exists.
>Equipment that just need power, according to .3af  should not be disturbed
>by the SYNC pulse Only  equipment that need SYNC or Power and SYNC will use
>it. Other equipment are  not affected at all   By  doing this, you bring
>new requirements on 802.3af. This might be an extension of  the existing
>PAR or a  new PAR. But at least, initial discussion should take place in 
>.3af   Raymond    -----Message d'origine-----
> : Bob Bell    [mailto:rtbell@xxxxxxxxx]
> : mardi 31 juillet 2001    22:32
> : 'Geoff Thompson'; 'Tony Bayley'; 'Raymond.gass';    'WM'
> : stds-802-3-pwrviamdi@xxxxxxxx; 'Tony Jeffree';    bob.grow@xxxxxxxxx;
>davel@xxxxxxxxxxxx;    David_Law@xxxxxxxxxxxxxx
> : RE: [802.3af] clocking over    Ethernet
>    -    If I    could speak from experience, IEEE 802.9b provided for a
>timing signal embedded    in the structure of the information flow. Even
>though we developed the    standards to a degree that was very
>implementable, the necessity of opening up    everyone's PC and installing
>a new NIC card made the transition to the 802.9    technology unsalable. If
>you truly want to pursue that approach, reactivating    802.9 might be the
>way to continue.       Bob    Bell        -----Original Message-----
>From:      owner-stds-802-3-pwrviamdi@xxxxxxxxxxxxxxxxxx     
>[mailto:owner-stds-802-3-pwrviamdi@xxxxxxxxxxxxxxxxxx]On Behalf Of     
>Geoff Thompson
>Sent: Tuesday, 31 July, 2001 11:01      AM
>To: Tony Bayley; Raymond.gass; 'WM'
>Cc:      stds-802-3-pwrviamdi@xxxxxxxx; Tony Jeffree; bob.grow@xxxxxxxxx;  
>   davel@xxxxxxxxxxxx; David_Law@xxxxxxxxxxxxxx
>Subject: RE:      [802.3af] clocking over Ethernet
>The      work of this group is limited by the "Scope" and "Purpose" of the
>PAR for      P802.3af:
>6. Scope of Proposed Project
>Define methodology for the provision      of power via balanced cabling to
>connected Data Terminal Equipment with      802.3 interfaces. The amount of
>will be limited by cabling physics      and regulatory considerations.
>with existing equipment will      be considered.
>7. Purpose of Proposed Project:
>To provide power for a      new class of devices with 802.3 interfaces
>enabled by
>progress in silicon      technology. These devices are characterized by low
>requirements and      LAN      connectivity.
>Your      proposals go well beyond these limits and, rather than being a
>benign add-on      to existing twisted-pair Ethernet, would be a radical
>change to the entire      asynchronous nature of Ethernet as it currently
>exists. It is very clearly      not an appropriate addition to the current
>project. for P802.3af which is      about power.
>If you feel that trying to add synchronous timing to      Ethernet networks
>is an appropriate proposal for an amendment to the IEEE      802.3 Standard
>then the proper method for pursuing this is found      at:
>See section 7.1
>At best, such a project      would only give you timing control within
>Ethernet links and across Ethernet      MACs. The transit time of
>information across Ethernet switches and routers      is completely outside
>the scope of IEEE 802.3. 
>Please move this      discussion to another venue.
>Geoff      Thompson
>                                |
>                                    |
>                    |
>                          |
>                                     |
>                      |
>                              |
>                                |
>                           |
>          |
>  |
>At      02:56 PM 7/31/01 +0100, Tony Bayley wrote:
>     I would also like a method of achieving        synchronisation to a few
>  However The        power supply in power over MDI
>  Typically only the
>  The
>hubs, switches, etc nearer        the network core are unlikely to be powered
>via MDI, so the        network-edge powered-devices will be in discrete
>clustered        around their own PSE, with no universal power over MDI
>  Timing information distributed along with
>power        over MDI, in this scenario, would only allow synchronisation of
>the PDs        attached to a single PSE, and would not be scalable to      
> larger
>Tony Bayley
>Principal Hardware        Engineer
>Bluesocket        Limited
>Jellicoe House
>Botleigh Grange Office Campus
>Grange        Drive
>Hedge End
>SO30 2AF
>United        Kingdom
>  tbayley@xxxxxxxxxxxxxx
>  +44 (0)        1489 773983
>  +44 (0) 1489 773966
>-----Original        Message-----
>From: owner-stds-802-3-pwrviamdi@xxxxxxxxxxxxxxxxxx
>[mailto:owner-stds-802-3-pwrviamdi@xxxxxxxxxxxxxxxxxx]        On Behalf Of
>Raymond Gass
>Sent: 31 July 2001 08:47
>To: Steve        Jackson; WM
>Cc: stds-802-3-pwrviamdi@xxxxxxxx
>Subject: RE: [802.3af]        clocking over ethernet
>the problem I see with NTP is        jitter and wander that makes
>it not really appropriate to generate a        sync pulse
>I like the proposal from Joerg; should be        investigated
>-----Message        d'origine-----
>De :        owner-stds-802-3-pwrviamdi@xxxxxxxxxxxxxxxxxx
>[mailto:owner-stds-802-3-pwrviamdi@xxxxxxxxxxxxxxxxxx]De        la part de
>Steve Jackson Envoy : lundi 30 juillet 2001 14:49  : WM Cc        :
>Objet : Re: [802.3af] clocking over        ethernet
>Your application may be best served by using        the Network Time Protocol
>It is fairly inexpensive to        implement a small NTP client in Ethernet
>devices, and getting        cheaper.
>You may wish to look at this comprehensive URL:
>and also read        the archives at the Usenet        newsgroup:
>        comp.protocols.time.ntp
>Steve        Jackson
>* * *
>WM wrote:
>> Hello        together,
>> sorry for dissturbing. I know that this is maybe        not the right
>> to ask
>> for it but maybe here are the        right people to give me a helpful 
>> answer. In principle it could        be also a nice add on for 802.3af.
>> Power over MDI is what        we are looking for for several years and
it is 
>> a must for several        applications like VoIP phones etc. but to fully 
>> utilize and        replace ATM, ISDN or other media by ethernet I think at 
>> least for        migration a way to synchronize devices would be
usefull. An 
>>        independent clock source like GPS is not usefull in our
>> Is there any way to clock or at least sync devices on the        ethernet? 
>> One way
>> would be to use e.g. the link pulses        but switches dont guarantee
>> all ports are in sync. The power        supply in power over MDI would
>> be a way to add a sync        pulse.
>> Many thanks in advance for your comments.
>>        Best regards
>> Joerg
>>        --------------------------------------------------
>> Wilhelm        Mikroelektronik GmbH
>>               Germany
>>    fax: ++49 2306        928289
>>      --------------------------------------------------

Wilhelm Mikroelektronik GmbH
   Sueggelstr. 31  -  44532 Luenen  -  Germany
  phone ++49 2306 928280  fax: ++49 2306 928289