<<   <   Page 5 / 5
Balet 12/5/2012 | 1:23:15 AM
re: Polatis Plugs On With Optical Switches Hi Arch_1,

Hopefully you have nothing to do with Arch Ventures:)

You are right on the first point.
You are also right on the scoend point.

There are a few other applications for switches, right? Protection, for example.
Switching ROADMs, hopefully much more often than once a month.

The perfectly valid paradigm for universal OOO network is a number of optical start-ups hungry to make big buck. I don't know any other yet.
Raymand 12/5/2012 | 1:23:07 AM
re: Polatis Plugs On With Optical Switches arch_1,

You are correct that the nonexistance of optical RAM, ORAM, for buffering data while headers are routed at intermediate nodes is the primary hurdle facing OOO statistical multiplexing. This is the biggest hurdle facing optical packet switching, OPS and I don't see that it will be solved for 10 years (and even then, not economically).

The reason I mentioned optical burst switching, OBS, however is because it avoids the need for ORAM by separating and aggregating headers onto a single OEO channel while transmitting and statistically multiplexing data payloads in an all OOO manner.

I am very aware that as switching times exceed 100 ns, that the overall efficiency of statistically multiplexing information on that channel begins to decrease (assuming a 10Gbps data rate). Because of this only electro-optic based switches will ever be suitable for OOO statistical multiplexing (regardles of if you are using an OPS or OBS approach). MEMS, piezo, or any similarly slow technology will be only be suitable for at most dynamic circuit switching, i.e. when the duration of the circuit greatly exceeds the duration of a specific data transmission.

There are fast switching technologies which are suitable even if they are not yet economical:

1. SOA - Solid state optical amplifiers are integrable and they switch in 1 - 10 ns. Their noise contribution however (2x an EDFA) which in large switches limit the physical path length since an optical signal can only go through 60 - 90 amplifications before requiring 3R regeneration (an OEO operation). If we assume switching fabrics are 32x32 crossbar's, then you would be able to traverse at most two intermediate nodes with the third node having to be the destination (96 SOA's even if the SOA has 1X noise).

2. TIR Switching - Total internal reflection switching can be achieved by the use of nonlinear optical materials which also switch in 1 - 10ns. Optical PLZT could be used to make integrated optic TIR switching arrays if coating chambers capable of making thick blanket films of optical grade PLZT were available. TIR switching does not introduce noise into the data (since it is simply a mirror and does not have any spontaenous emission components). TIR switching could easily support large crossbar switching fabrics without degrading the path length of a networks made from such nodes, i.e. you could maintain your 3000 km optical radius (absent coupling losses) support a large number of even large fabric intermediate nodes.

There is debate on whether data flows at the networks are still bursty or whether they are statitically smooth. Certainly today they would be smoother due to the amount of statistical multiplexing that is occuring through aggregation up to a core data rate. Whats not certain however is whether the core will become more bursty as larger data payloads (such as a multi-Gb movie download) become demanded by end users. I think the core will become increasingly bursty as end users gain the ability to consume data in large chunks as a natural outfall of consumer choices to download a movie or video link to their grandmother (still some years from now).

In the end however, I am driven back to my original question as to if an when even statistically multiplexed OOO solutions exhibt a sustained economic benefit relative to OEO approaches.


Everybody Loves Raymand

Raymand 12/5/2012 | 1:23:06 AM
re: Polatis Plugs On With Optical Switches arch_1

A second item that I missed the first time.

I noted that you identified TDMA Star Switching as("burst switching"). This is not what is ment by the term "optical burst switching" or OBS.

OBS uses the electronic buffer memory at the edge of the network as payload buffer while the header for a burst traverses the network along a control packet wavelength in a normal hop-by-hop IP manner. The header enters the network first while the payload waits for an offset time at the edge. At each node, the header (also called a control packet) makes a "future" reservation for the payload on a data wavelength lying along the data (i.e. MPLS) path. After the offset time elapses, the data payload is launched into the network along the MPLS/data wavelength path. Because each node was told by the header when and on which wavelength the payload will arrive, each node configures its switching fabric just before the payload arrives so that the payload traverses the node from the known input to the reserved output. Once the payload has traversed a node, the switch fabric is free to immediately reconfigure for the next payload due to arrive

Note: If the fabric is non-blocking then multiple payloads would be simultaneously traversing the node on independent paths without conflict unless two payloads were asking for the same exit path and wavelength wherein a contention resolution scheme would kick in.

Theoretical studies on OBS indicate that this reservationless (i.e. payloads are launched without hearing back any acknowledgements that reservations have been successfully made), bufferless networking schema is able to carry higher loads (or equivalently exhibit lower rates of packet loss) than OEO IP networks.

OBS is -- A New Switching Paradigm.

New paradigm aside, is it cheap enough to provide a sustained economic incentive to adopt? At very high demands of high speed bursty data to a high number of end users (nationwide population of Gb+ data requests), OBS may well become sustainably cheaper since its ability to efficiently scale IP centric capacity at very low incremental cost may become the tipping factor.
Balet 12/5/2012 | 1:23:05 AM
re: Polatis Plugs On With Optical Switches Raymond wrote: There are fast switching technologies which are suitable even if they are not yet economical:

You seem to know the future well.
And you are tight on SOA and PLZT switches.
I was told by an expert that PLZT might be too difficult technology to ever become manufacturable, inspite of young startups coming up promoting it.

Bell labs group is doing a nice job on fast swtiching. They seem to be progressing fast (unusual for Lucent).
venzz_z 12/5/2012 | 1:22:35 AM
re: Polatis Plugs On With Optical Switches I'm not really familiar with PZT technologies, but as far as I know, about SOA switching, I don't think it is to come as a plug and play product for switching quite soon yet.

There's been many switching configurations that were investigated for switching with SOA. For interferometric configurations, some of the most notable ones are probably the Symmetric Mach-Zehnder, the SLALOM / TOAD configuration or the UNI. There are some other switches that use the cross-gain modulation of the SOA to either switch or do wavelength conversion or other physical mechanisms.

But for all the configurations I've seen in the past years, they all are limited in switching speed by the carrier relaxation time to the order of ns. Electronics work quite well for up to 10G, so SOA switching under 10G competes with electronics. Electronics weak point is power consumption, so I'm not sure if SOA will become competitive under 10G.

For switching speeds higher than 10G, it seems that it can be done with SOA, but I believe the integration technology to get low coupling loss is a major factor. Loss means amplifier, which means noise. Therefore, the power penalty might not be reasonable for a network with many interconnections but for the crossbar of a fabric, its probably not impossible.

As for 3R-OOO, I don't believe there's going to be a great breakthrough any soon. There's been a long time people are talking about OOO but unless a great material with high nonlinearity, integration capabilities and low loss comes up, its not for any soon.
<<   <   Page 5 / 5
Sign In