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


  I would not agree with your description that Stat Muxing and
DBA are two different things.

Stat Muxing can happen at different levels.
The traffic from the behind the ONU (could be a LAN) is
Stat Muxed in the ONU. This level of stat Muxing is obtained
both by SBA (static bandwidth allocation) and DBA.
The next level of possible Stat Muxing is between traffic from
different ONUs. This can be obtained only by DBA. Our presentation
and our traffic studies show that Stat Muxing between ONUs is
important because even the data traffic after the first level of stat
muxing is very bursty. As you can see in our presentation, simple DBA
can provide 20-30% improvement (based
on traffic type) over SBA schemes. I would further like to note that
the efficiency numbers you show in your presentation can only
be achieved only theoritically at infinite delay. It will be useful
to the group show the delay of a TDMA based scheme as a function of
It seems to me that there is a consensus in this group that
dynamic adaptation is required though there are differences on how
exactly and on what time scale the adaptation is done.

I also heard a comment in this thread that QoS is not within the
scope of the project. Subscriber access is a new space for Ethernet.
Ethernet never tried to support voice, video and data. It is necessary
provide mechanisms in the EFM standard to provide some level of
QoS support. I will try to make the case with an example. If a
Voice packet arrives behind a large data packets, mechanisms to transmit
voice packet ahead of the data packet will be useful. The arguement
presented in this thread is that this should be performed above the
MAC. How will a higher layer entity know the size and interval of the
time slot for this ONU? How can one control the timing between a
physical layer running at Gbps and a high layer running on a different
clocks? These interfaces are important if different vendors are to
implement different parts of the overall system and they are expected
to interoperate. While we do not want to burden the standard with
ATM like QoS we need some "lite" mechanisms to differentiate service.

-----Original Message-----
[]On Behalf Of
Sent: Saturday, July 14, 2001 2:33 AM
To: zhangxu72@xxxxxxxxx; RHirth@xxxxxxxxxxxx
Subject: RE: [EFM] RE: EPON TDMA

These are comments for both Xu's and Ryan's postings.

First let's not mix stat muxing and dynamic bandwidth allocation. These are
different concepts.

DBA is a method allowing "just-in-time" bandwidth allocation to an
application that requires it.  As an example, consider a network carrying
voice and data.  In the absence of voice traffic all the bandwidth is given
to data traffic.  When new voice call arrives "some mechanisms" will reduce
the bandwidth available to data traffic and will allocate it to voice
traffic.  This bandwidth will be guaranteed to voice traffic in a sense that
each voice packet won't need to struggle to get its share of the bandwidth.
When voice call completes, the same "mechanisms" will return the bandwidth
back to data traffic.

Statistical multiplexing is a way of statistically allocating channel
bandwidth, i.e., stealing chunks of bandwidth when other users (node) failed
to do so.  "Statistical" nature means that bandwidth available to a user
will converge to some fixed value only when averaged over long observation
time.  But there is no way to predict how much bandwidth will be available
to a node in any given short interval of time.

Ethernet (specifically CSMA/CD) uses statistical multiplexing. DBA, on the
other hand, was never part of Ethernet.  But when we think of Ethernet in
the First Mile, we realize that this is whole new world for the Ethernet,
where it has never gone before.  Suddenly stat muxing in its current form
(CSMA/CD) becomes very harmful due to its statistical nature.  Yes, we want
to utilize bandwidth efficiently, but most importantly - we need to provide
SLAs to users.  CSMA/CD is a non-deterministic service: packets may collide
some number of times and then be discarded. DBA in this new world becomes
important, as we want to be able to deliver all services: voice, video,
data, etc.

How this could be solved in EFM?  Let's first consider P2P solution as I see
it.  In P2P deployment a very smart switch will be located in CO.  This
switch will monitor traffic for each user in terms of both volume and
application.  As the uplink bandwidth is clearly a limited resource, the
switch will make an arbitration decision of which packets to drop in terms
of both keeping the user within its pipe and maintaining some sort of DBA
within each pipe.  We hope the switch will be SLA-aware.  Of course, it will
be proprietary to each vendor how switch fabric will be implemented.  It is
higher level, above MAC and PHY, and the standard is not concerned with it.
The point is that both decision of how to keep user within its pipe and
execution of this decision are done in the CO.

Now consider P2MP.  In the same way as in P2P, higher layers in OLT will
make a decision how to keep user within its SLA.  The only difference is
that execution of this decision and ensuring DBA within user's pipe are
delegated to an ONU.  And if in P2P the switch may decide to give entire
uplink bandwidth to one ONU, so in P2MP, the OLT may do so by giving all
timeslots to one ONU, or just by making it one large timeslot.  Of course,
real implementation is a bit more complicated: changed ONU state needs to be
propagated to OLT. This may be done through OAM communication channels,
proactively of otherwise, and except increased delay has no side effects.
Letting PHY be timeslot-aware is just a mechanism for ONU to execute the
OLT's decision.  OLT may choose to modify timeslot assignments or size as
often as it deems feasible. Specific values of timeslot, frequency of
updates, and algorithm used to make such decisions are all outside the scope
of the project.

I readily agree with Xu's comment that we need a model to analyze. Once EFM
graduates into a work group and technical work begins, I think we will
proceed by building a simulation model for various approaches.

On a general note, I would like to suggest to group members to refrain from
comments like "Ethernet never did that..." or "Ethernet traditionally does
that...".  Ethernet traditionally supported CSMA/CD, and in 802.3ae it
doesn't anymore.  And it never was used in WAN and now it is.  Ethernet
never had OAM, and now it will. Without fair amount of "heresy" in each new
project Ethernet would never become ubiquitous protocol as it is now.  We
have PAR and objectives to govern our direction. Tradition and religion is
not one of them.

Thank you,


-----Original Message-----
From: xu zhang [mailto:zhangxu72@xxxxxxxxx]
Sent: Friday, July 13, 2001 7:28 PM
To: Ryan Hirth
Subject: RE: [EFM] RE: EPON TDMA

I agree with Hirth's opinion, in order to keep the
statistic multiplexing nature of ethernet, the DBA
is needed.
in a large time solt. such as 125us, if the ONU has
large traffic, the time solt may be not enough, if the
ONU has little traffic, the bandwidth utilization will
be reduced a lot. In a fixed size time solt, the DBA
is easy to implement, but in order to achieve high
bandwidth utilization the time solt need to be small,
when using variable size time solt, the DBA is hard to
implement, but it can keep  statistic multiplexing
nature of ethernet and at the same time achieve high
bandwidth utilization.

I think whether the frame will be segmented of not
segmented, how long the time solt will be,
the DBA or SBA(static bandwidth allocate£(c)£¨
using variable size time slot or fixed size time slot,
we need a model to calculate.

--- Ryan Hirth <RHirth@xxxxxxxxxxxx> wrote:
> Ethernet has always had an inherent form of DBA in
> the fact it allows a
> station with traffic to send at up to the line rate
> or an arbitrated rate
> less than that.  However in a connectionless system
> there are no service
> contracts or allocations of that bandwidth, but
> bandwidth of the media is
> divided dynamically.  SLAs are features which do not
> belong in the Ethernet
> MAC layer, however dynamic bandwidth allocation is
> inherent within Ethernet
> and that is why Ethernet is so well suited for data
> traffic.
> By creating fixed timeslots in the upstream you are
> changing the nature of
> Ethernet.  Now the maximum bit rate of one station
> to burst upstream is
> limited to its timeslot.  I believe according to the
> AllOptic presentation
> this would be 25 - 50 Mbps/ station (without DBA).
> This creates asymmetry
> which has never been an explicit form of Ethernet.
> A new media for Ethernet should present similar
> characteristics of
> traditional Ethernets.  There is certain level of
> service which Ethernet
> has.  If you increase the latencies across the media
> ten fold, is it still
> Ethernet?  The end user will perceive a difference
> in service.
> Ryan Hirth
> Terawave Communications
> rhirth@xxxxxxxxxxxx
> (707)769-6311
> -----Original Message-----
> From: jc.kuo@xxxxxxxxxxxx
> [mailto:jc.kuo@xxxxxxxxxxxx]
> Sent: Friday, July 13, 2001 4:06 PM
> To: glen.kramer@xxxxxxxxxxxx; zhangxu72@xxxxxxxxx
> Cc:
> Subject: RE: [EFM] RE: EPON TDMA
> As PON is just a new media of Ethernet, the overall
> system will be a base on
> "Switched Ethernet" architecture.
> Under this architecture, bandwidth shaping and
> priority queuing will only be
> done in the switch fabric. In the MAC and PHY, a
> mechanism which allow
> flexibly assign the data rate may benefit the DBA
> implementation but DBA
> algorithm will not be implemented as part of MAC and
> PHY layer function.
> There is always trade-offs between delay and
> utilization. Reduce the guard
> band and do the packet fragmentation will help the
> bandwidth utilization,
> then the delay can be minimized. EPON is under the
> umbrella of Ethernet,
> keep the Ethernet frame integrity is one of the
> religions of 802.3 team,
> packet fragmentation is not considered as an option
> for the standard.
> JC Kuo
> jc.kuo@xxxxxxxxxxxx
> Alloptic, Inc.
> 2301 Armstrong St.
> Livermore, CA 94550
> Phone: (925) 245-7641
> Fax: (925) 245-7601
> -----Original Message-----
> From: glen.kramer@xxxxxxxxxxxx
> [mailto:glen.kramer@xxxxxxxxxxxx]
> Sent: Friday, July 13, 2001 2:55 PM
> To: zhangxu72@xxxxxxxxx; glen.kramer@xxxxxxxxxxxx
> Cc:
> Subject: [EFM] RE: EPON TDMA
> Dear Xu,
> I think I know what confused you in the presentation
> as I got several
> similar questions.
> Timeslot is not an analog to a cell. While, from the
> slide 4 in the
> presentation you may conclude that one timeslot is
> only large enough to hold
> one maximum size packet, that is not the case.
> Timeslot in our example was
> 125 us, which equals to 15625 byte times.  Then you
> can see that in the
> worst case it will have 1518 + 4(VLAN) +
> 8(preamble)+12(IPG) - 1 = 1541
> bytes of unused space at the end of timeslot
> (assuming there is data to be
> sent and no fragmentation).  With realistic packet
> size distribution (like
> the one presented by Broadcom), the average unused
> portion of the timeslot
> is only about 570 bytes.  That gives channel
> efficiency of 96%, or
> accounting for 8 us guard bands - 90%
> DBA is a separate question.  While it may be
> important for an ISP to have
> DBA capabilities in their system, I believe it will
> not be part of the 802.3
> standard.  But a good solution would provide
> mechanisms for equipment
> vendors to implement DBA.  These mechanisms may
> include, for example, an
> ability to assign multiple timeslots to one ONU or
> to have timeslot of
> variable size. Grant/Request approach is trying to
> achieve the same by
> having variable grant size.
> Having small timeslots will not solve QOS either.
> Breaking packet into
> fixed small segments allows efficient memory access
> and a cut-through
> operation of a switch where small packets are not
> blocked behind the long
> ones (and it assumes that short packets have higher
> QOS requirements).  In
> such a distributed system as EFM is trying to
> address (distances in excess
> of 10 km) the gain of cutting through is negligible
> comparing to propagation
> delay or even the time interval before ONU can
> transmit in a time-sharing
> access mode (be that TDMA or grant/request method).
> Thank you,
> Glen
> -----Original Message-----
> From: xu zhang [mailto:zhangxu72@xxxxxxxxx]
> Sent: Thursday, July 12, 2001 7:01 PM
> To: glen.kramer@xxxxxxxxxxxx
> Cc:
> Subject: EPON TDMA
> hi, glen:
>  I had seen your presentation file about EPON TDMA
> in
> PHY, it help me a lot to understand your EPON
> system.
> We had developed the first APON system in china,
> when
> I think of the TDMA of EPON, I think though the
> uplink
> data rate is 1Gbits/s when shared by 16 or 32 users
> is
> still not enough, so the dynamic bandwidth
> allocate(DBA) protocal must be a requiremant
> especially when take care of the QoS performance. In
> DBA protocal, in order to achieve high performance
> the
> time slot need be to small, I think why not we
> divide
> the ethernet packet to 64 byte per solt, it is often
> used in ethernet switch when store packet in SRAM.
> best regards
> xu zhang
> __________________________________________________
> Do You Yahoo!?
> Get personalized email addresses from Yahoo! Mail

Do You Yahoo!?
Get personalized email addresses from Yahoo! Mail