Cut Through Routing
Return of an old idea?
If you do a Google search on “Cut through routing”, you will get 24,400 hits. Refining it to “Cut through routing” and “optical” results in 684 hits. “Cut through routing” and “ATM” results in 742 hits. Thus, it’s not surprising that if you refer to cut-through routing, many people will think of ATM networks. That is not a bad starting point since work on using cut-through routing in optical networks is focused on using Photonic Crossconnects (PXCs) in the center of IP-routed networks to reduce costs and improve network performance just as it was in IP/ATM networks.
There are several approaches to mixing IP routers and PXCs with the major differences being in how the decisions to utilize the photonic paths are made and the scope of those decisions. A major question is how to recognize which data streams (flows, bursts, etc.) should be directed through the PXCs. For example, an email transmission might not benefit from the higher quality photonic path, a large file transmission could benefit by reducing the load on the IP routers/switches and a streaming video might require the higher quality path. In this article, I will briefly describe a few of the approaches.
Edge Router Controlled Photonic Networks – The approach of using edge routers to control a network of PXCs in the center dates back to the origins of GMPLS. The common phrase was “Smart edge, dumb core”. As manufacturers of PXCs, we object to the term “dumb”, but concept is solid. Intelligence belongs at the edge where the traffic is aggregated and electronics is already looking at the bits. Once the traffic is aggregated, there should be no need to convert back to electrons (except when required by the laws of physics, of course).
Central Control of Photonic Networks - Work on this approach has been done by NTT. Their network approach is similar to that of the edge router controlled approach, a mixture of routers and PXCs, but it does not use edge control. Instead a separate monitoring entity continually monitors all parts of the network. Analysis of traffic patterns leads to decisions about what traffic should be directed through the PXCs and how the PXCs should be used to reconfigure the network for current traffic loads. For more information on the NTT work, see: 'Cut-through Optical Path Control Technology for a Terabit-class Super-network, Kenichi Matsui, Takeshi Yagi, Yuuichi Naruse, and Junichi Murayama, NTT Information Sharing Platform Laboratories' http://www.ntt.co.jp/tr/0403/papers.html
Both of the above approaches treat the IP routers and the PXCs as separate network elements. Another path to cut-through routing leads through more tightly coupling a PXC with an IP router creating a hybrid router. There are two branches in this path.
Hybrid Router with End-to-End Paths – An example of this approach is the Hikari router prototype built and demonstrated by NTT in 2000. As shown in the following diagram from NTT, GMPLS was used to create lightpaths which either traveled end-to-end in the optical domain, or were terminated in the electronic part of the router. As such, this approach can be viewed as expanding the traffic handling capacity of the router. As in the case of ROADMs, through traffic which does not require the electronic capabilities of the router are directed to the pure optical path. This frees up router capacity to handle other traffic. In the case of Hikari, NTT was able to show Tb/s capacity. For more information on Hikari, see: ' "HIKARI" Router - Core Node System for Next Generation Photonic Backbone Networks' http://www.onlab.ntt.co.jp/en/pt/router/index.html.
Hybrid Router with Local Paths – This is an interesting approach
being pursued by CTVR at Trinity College in Dublin. In order to simplify
the decision making and to ease network insertion, decisions in this approach
are made purely locally and implemented through negotiations only with
nearest neighbors. The router analyzes its own traffic and decides what
traffic should be directed through the PXC section. Network elements on
both sides are then instructed to direct the traffic to the pure photonic
path. Thus, the traffic may be in the optical domain only for this node,
or the adjacent network elements may extend the photonic path if they
have PXC capability and decide independently that it should be extended.
The objective here is the same as in the end-to-end case, a significant
expansion of the capacity of the electronic IP router. For more information
on the CTVR work, see: Ruffini, M., O'Mahony, D., Doyle, L.E., 'A Testbed
Demonstrating Optical IP Switching in Disaggregated Network Architectures'
in Proceedings of the IEEE TridentCom, Barcelona, March 2006.
