Well, there's some optical. The point is that TeraStream is taking IPoDWDM seriously and might be a showcase for Cisco's homemade transceivers

Craig Matsumoto, Editor-in-Chief, Light Reading

December 17, 2012

4 Min Read
DT's All-Cisco, No-Optical Network

The new Deutsche Telekom AG (NYSE: DT) architecture, TeraStream, skips using optical equipment and might be using a new Cisco Systems Inc. (Nasdaq: CSCO) 100Gbit/s optical interface, according to a description from a source familiar with the specifics.

DT is also taking advantage of the IPv6 header for storing policy-related information, thus letting services assign policy based on a quick glance at the header.

It's no surprise that TeraStream would try some newfangled ideas. That was the point, after all. TeraStream is a modernized architecture that runs on IPv6 only -- no ATM or Frame Relay -- and exploits cloud-networking concepts. All services in the network are handled by virtual appliances in the data center, an arrangement that smacks of software-defined networking (SDN).

The first TeraStream rendition is running in Croatia, where Hrvatski Telekom, a DT subsidiary, is using it to deliver residential services. The pilot went live on Dec. 10. (See DT's Croatia Unit Taps Cisco.)

Cisco, which is supplying the TeraStream gear, made that announcement and detailed quite a bit about TeraStream's inner workings. For instance, Cisco is supplying the data centers for TeraStream, basing them on the Unified Computing System (UCS) architecture.

But our source, who didn't want to be named, seems to think the finer points of TeraStream have gone unappreciated. So let's take a look at what's under the hood.

No Optics Allowed (Almost)
TeraStream goes whole hog with IP over DWDM (IPoDWDM), meaning the optical transceivers are directly on the routers, eliminating the need for transponder shelves. There are no optical subsystems in the network at all, save for passive splitters and EDFAs, according to the source.

Those IPoDWDM interfaces apparently run at 100 Gbit/s, which brings up the possibility that they're based on Cisco's new, homemade transceiver format called the CPAK. It's a smaller alternative to the CFP, and it's come out earlier than the CFP2, the "official" smaller-than-CFP format.

Cisco hasn't officially acknowledged the CPAK, but as of November, sources inside the company were telling Light Reading its launch was imminent.

If CPAK isn't already in TeraStream, it's probably going to get there soon; the source says DT is hoping the use of silicon photonics will bring 100Gbit/s and 400Gbit/s pluggable transceivers below the $10,000 level. CPAK is assumed to be using silicon photonics that Cisco acquired with Lightwire. (See Lightwire Points Cisco Toward 100G and Silicon Photonics Signals Red Alert for 100G .)

The TeraStream architecture was drawn up by Peter Löthberg and Guenter Honisch, the source speculates. Honisch is part of DT's office of the CTO, while Löthberg is an optical networking legend known for, among other things, bringing a 40Gbit/s connection to his mom's house.

Here's the full text of what Light Reading was sent last week. We've broken it into paragraphs for readability.

Terastream, believed to be the brainchild of Lothberg & Honisch, completely integrates the whole carrier network and datacenter.

The network is designed with only two layers of routers. One set of routers are used for customer aggregation and all policy, and the other to combine the functions for core, peering and datacenter switching/interconnect. The internal network is IPv6 only.

Traffic type is encoded into the IPv6 addresses, allowing policy processing for different services through a single ip lookup (eg. labels moved to the addresses).

All the transport links in the network use 100G coherent optics tightly integrated with the routers. One can guess Lothberg is behind this and some information indicates that he is working on the assumption that silicon-optics will be able to make pluggable 100G/400G transceiver for less than $10K. The design has removed all optical subsystems and reduced the whole transport to passive splitters and EDFA amplifiers in a completely colorless drop-and waste model.

All protection is IPv6 based. QOS model is turned around with bidirectional policing at the edges.

Two modes of operation are provided, one IPv6 native and the other a Metro-Ethernet MEF compliant customer facing service utilizing IPv6 transport encapsulation and full Y.1731 support.

By using IPv6 the design can interconnect a network endpoint directly to an application running on a VM in the cloud datacenter, IPv6 also allows direct addressing of a logical port, rather than having content context in the encapsulation header.

All traditional network services are moved from the network to the network-centric datacenters. In a presentation done by Honisch he said the services were provided by "virtual appliances" and indicated that "buying appliances and service blades as a software license."



— Craig Matsumoto, Managing Editor, Light Reading

About the Author(s)

Craig Matsumoto

Editor-in-Chief, Light Reading

Yes, THAT Craig Matsumoto – who used to be at Light Reading from 2002 until 2013 and then went away and did other stuff and now HE'S BACK! As Editor-in-Chief. Go Craig!!

Subscribe and receive the latest news from the industry.
Join 62,000+ members. Yes it's completely free.

You May Also Like