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

Fiber compatability - another historical reminder


While there is no question at first glance that it is desireable to develop
a standard that takes into account the installed base, I think it may be
useful to remember the work that was done in this very group on the
100Base-T standard only a few years ago. In the interest of supporting the
vast installed base of CAT-3 wiring (which was the required wiring for
10Base-T), the IEEE 802.3 group decided to develop TWO standards for copper
based 100Base-T.

100Base-T4 was similar to the WWDM proposal for 10Gbps Ethernet in that it
seemed at the time to be the simplest method to support the huge installed
base (tens of millions of nodes!) of 10Base-T users who had CAT-3 wiring.
Similar to WWDM it contained four "lower speed" transmitters. The "other"
flavor, 100Base-TX was targetted at people who had the newer and higher
quality CAT-5 cabling. 100Base-TX has a single faster transmitter (similar
to the serial 10Gbps). Many respectable companies developed transcievers to
support both standards (usually separate chip sets).

The result? 4 years later 99% of the copper 100Base-T market today uses
100BaseTX. The vast majority of people ended up biting the bullet and
changing their cabling. 100Base-T4 product lines have been discontinued by
most companies.

A number of observations:

- In the case of 100Base-T4 and 100Base-TX the consideration was compability
with very complex wiring that connects to every desktop in the office.
Wiring which is usually embedded in walls and in very complex
configurations. In the case of 10Gbps Ethernet, the fiber concerned will
usually be fairly well contained and easily identifable (connection to
servers or between switches, very few desktop connections). Regardless of
cost, for somebody who really wants 10Gbps, how much of a barrier is it to
replace the small number of fiber links? Before making a final decision on
the importance of cable compability, the true task, complexity, and cost of
replacing this wiring should be measured against the long term potential
benefits of defining a standard which may be optimized for a better long
term solution.

- In retrospect, it turns out that since 100BaseTX only has one transmitter
(albiet a much faster one), it is able to offer a much more cost effective
solution than 100BaseT4 with its four transmitters. This was not obvious at
the time the standards were being defined. We need to be careful not to fall
into a short-term cost benefit trap. Although WWDM appears to be a more cost
effective solution today, history would lead us to believe that in the
longer term, a single transmitter approach may be the better way to go.
Technological problems which seem signficant today have a way of getting
solved with advances in technology and with clever engineering.

It is difficult to argue with a blanket statement that backward compability
is good, but discussing this seriously today will hopefully result in the
truely correct decision as we move forward.

Pini Lozowick
Chiaro Networks

Thomas Dineen wrote:

> GentlePeople:
>     As an active participant in recent 803.3 affairs and study groups
> there is one thing I have learned. The IEEE is a conservative
> organization
> which is extremely bound by rules and traditions, or at least this was
> the
> lament which seemed to prevail.
>    So some conservative ideas and principles:
>    1) IEEE 802.3 is a member group within IEEE 802. The charter of IEEE
> 802
> is LANs and MANs and NOT WANs. The work of IEEE 802 has been directed,
> so far, exclusively to the LAN and MAN arena. The the first conservative
> principle I would like to espouse is that 802.3 should NOT attempt to
> venture
> into WAN Standards. The 802.3 HSSG study group should devote itself
> exclusively
> to 10GBit/Sec LAN and MAN Standard development. I believe to do
> otherwise would
> violate the IEEE charter and principles, not to mention long held and
> beloved
> conservative traditions. WAN standard development should be left to
> those who do
> it best, the accredited WAN Standards groups.
>    2) Now from a technical viewpoint. LAN/MAN standards and the
> protocols of
> which they are composed differ greatly from those of the WAN. This is
> justly
> based on the differing operating requirements of the applications
> involved.
> For example LAN/MAN Standards operate over relatively short distances,
> with very low bit error rates, in environments where the equipment and
> media
> are under private user control. In contrast traditional WAN environments
> operate
> over relatively long distances, many times with relatively high bit
> error rates,
> in environments where the equipment and media are under the control of
> one or
> more public network providers. The IEEE 802 LAN/MAN standards feature
> network
> management protocols, and the WAN Standards feature Operations and
> Maintenance
> Protocols. The bottom line is that these protocols and application
> spaces
> are vastly different and completely incompatible. So the second
> conservative
> principle I would like to espouse is that IEEE 802.3 HSSG should develop
> a 10GBit/Sec LAN/MAN standard with sufficient defined interfaces, such
> as
> a DecaGMII, so that traditional accredited WAN Standards groups may
> develop
> compatible PHYs which will support related WAN applications. In fact I
> would
> recommend that IEEE 802.3 request this via liaison letter.
> 3) Support for the installed base. A third conservative principle,
> we MUST support the LAN/MAN traditional installed base this includes
> both
> single mode and MULTIMODE fiber. Conservative principles, not to mention
> modern
> capitalism, demand this. I have heard the lament from some that there is
> difficulty supporting multimode fiber. Well over coming difficulties
> is the job of the study group, not to mention the reason you are paid
> the
> big bucks. Lastly what about copper support? If possible copper should
> be
> supported, although I could envision providing dispensations if it turns
> out that driving copper at these rates causes local warps in space and
> time
> continuum.
> Thomas Dineen