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

[EFM] Optical Switches for EFM PONS Systems

This is a mainly new thread developed from what has been previously talked
about on the EFM reflector and presentations given.  This idea uses most of
the ideas of EPONS with one perhaps slight modification to the word
"passive" in EPONS (ethernet passive optical networks).  In this thread
passive only means that the distribution line is passive in the optical
sense for the head end/Central Office to the end user link.  That is the end
users bits never undergo an OEO converstion en-route from the either the CO
to end user or the other way around.

In this thread I would like to introduce a new device called a
passive-optical switch.  The switch is passive in that the traffic which is
destined for another destination passes through the switch with out any OEO
converstions, but the switch itself is controled via electrical means.  Many
technologies are presently being worked on for this, including at the
company I work for.  I do not at this point want to state what technology we
are using because I don't want to get into any legal problems.   These
technologies should allow for under 100ns (nano seconds) switching times
between ports.

So long as switching can be done in under the time of the IPG of the
Ethernet Frame then the optical switches should be useful.  Since the IPG at
1 Gbe is around 88ns (not including the EPD byte).  Therefore these new
switching technologies look technically feasable.

I see such a switch as being able to address three of the challenges facing
a EPON based EFM system.
In the EFM sub area of EPONS discussion, particularly those using DBA
(dynamic bandwidth allocation), the following three issues seem to occur.
1) Scalability:  This problem actually would seem to occur both now and
later.  Presently sending even 1/32 of a 1 Gigabit stream to anyone person
seems a bit excessive.  Businesses may need such bandwidths, and even then I
would think only medium sized and larger enteprises.  In a FTTH (since this
is EFM we are indeed talking about how to bring ethernet to the end user,
i.e., not FTTC, but rather FTTH), even 10 Mbps would probably satisfy most
people needs (One could send a good quality video feed at this rate without
a problem).  Therefore the initial EPONS if deployed at 1 Gigabit will be
mainly unused.  The one caveat to this is if EPONS is ONLY used as an
overlay, not as a pipe for all the needs at the end user.  
 Today Cable Modems share a much smaller bandwidth with many more
subscribers and people who successfully use it seem to be happy.  On the
other end, once people are using the bandwidth, upgrading means
 a) Splitting the users into subgroups, by different fibers or WDM.  WDM
tends to be a bit costly and splitting into physically different subgroups
would be expensive if no spare fibers exist.
 b) upgrading everyone to a higher bitrate.  This would be expensive and
cumbersome to go to everyone's homes.  Especially since the use of bandwidth
will probably vary greatly between end users.

Therefore, it would seem to me that today in a non-overlay setting even a
1:32 split may be to little, but in the future it may be too much.  
2) Failure Containment:  Should one subscriber's transmitter go hey wire and
not turn off at the appropriate time, they will take down everyone and
tracing the problem will be very hard from the central office /head end.

3) Security:  Although snooping on someone else's conversation would be at
best a one way affair, it may be possible for someone to mimic someone
else's system and therefore have bits sent to them when we thought they were
being delivered elsewhere (spoofing).  This could lead to billing and other
security related nightmares.  I'm not saying this is likely, but something I
think EPONS needs to insure is addressed (either at the EPON 802.3 level, or
insure that a higher level can appropriately take care of it).  I understand
that some feel that this is an issue better addressed at higher layers, but
I include this for completness and because there may be some debate on this

With the use of optical switches the above three issues could be elegantly
solved.  First let me state in a bit more detail what I mean by an optical
 An optical switch is an nxm switch which switches optical beams from any
one of n-inputs to any one of m-outputs.   The switch can rearrange the
beams extremely fast (well within the time of the IPG).  Such switches while
not commercially available today are being worked on even as we speak and
some of them are designed to go up to very high port counts.
 How would this address the three issues raised above?  As follows...
1) Since the switch will have a constant loss for 16, 64, 256 or whatever
number of ports that can be established, the split among the various users
can be increased.  This will allow for better initial sharing of the
resources (before people need, on average, a throughout of 30 Mbps each).
If the area is filled with a businesses or a mix of businesses and
residential, or if the residents are all extreme surfers, then additional
input ports "m" could be opened.  They could use the same wavelength (if
 extra fibers are around between the optical switch and the head end) or
just two WDMs for everyone (one at the head end, the other before the "m"
input ports to the optical switch.  Of course now the DBA would need to
insure that no user gets a packet from two different wavelengths at the same
time (but adding this functionality should not be that much effort). The
receivers are wide band enough not to care about the wavelength being
2) Failure Containment would be easy in that if any one person started
transmitting in other peoples time slots the switch would prevent him from
effecting the others.  He would be detected as problematic because the head
end does not receive the packets (even say an I'm alive type packet) when
expected.  This could trigger an alarm that the CO engineer could then deal
 with, without interupting other paying subscribers.

3) Each end user would be connected to one of the "m" output ports.
Therefore since each user is switched his data and only his data, spoofing
someone can be made much more difficult.  
 Therefore by the use of a switch three major issues are resolved, and
although if we are only talking about 16 users, yes the passive splitter is
cheaper, if we consider that the number of users who share a single Gigabit
connection could be much higher (at least at first) then the use of a switch
by far more than 16 users may be quite justifiable.  Also the switch allows
for a much easier upgrade path.
 The inclusion of this architecture would really be just a generalization of
the EPONS architecture, mainly by the inclusion of WDM into the DBA, and the
ability to support more than 32 users at one time.
 Finally, although these switches don't exist in the market today, they are
under active development by a number of companies and by the time an EFM
standard is finished they may well be ready for market.
I look forward to hearing from people their thoughts on this issue.  
 Joshua Brickel
 email:  joshua.b@xxxxxxxxxx