Andreas HegersNovember 7, 2016
7:31 am PT

In 2012, an executive who worked for one of the largest European service providers shared his dream with the attendees of Layer123’s first software-defined networking (SDN) & OpenFlow World Congress in Darmstadt, Germany. According to him, the reason for all the misery in carrier-land were not permanent price battles, flat-rates, and cut-through competition; it was us, the vendors. We were accused of only offering hardware that was too expensive and wasn’t interoperable forcing carriers to pay much more than necessary. The audience was impressed — especially when we were asked to collaborate, define common interfaces, agree on one communication language, and offer open, fully programmable boxes based on standard industry architectures to drive his cost down as much as possible. For the majority of the audience, this might have been the first time when they were confronted with the realities of the new, open world. Since then, the world keeps on changing, but for some reason, both carriers and (most) vendors are still alive. But OpenFlow, the savior of that day, is virtually dead. What went wrong?

Well, quite a bit. It took a while, but people close to the matter realized that even after several massive iterations, OpenFlow would never be powerful enough for managing a dense wave division multiplexing (DWDM) system, let alone something as complex as a packet-optical transport system (P-OTS) or an optical transport network (OTN) switch. This didn’t stop some vendors from claiming that their transport solutions are fully compatible with this not so powerful language, so a lot of confusion was injected into our industry, but in reality it never worked. The idea of one powerful controller, to control all network items, only worked in theory since it would hamper operations substantially due to the readily available compute power and lack of common standards. Over time, the dream of one language, one central controller, and flat, programmable networks became a playing field for marketing departments and clueless journalists. SDN, announced as the new messiah, went down from being the Holy Grail to becoming the next version of network management systems (NMS). Element management systems (EMS’s) are called Domain Controllers, NMS’s are Parent Controllers and Umbrella Network Managements (remember those?) are called orchestrators.

Meanwhile network functions virtualization (NFV) had entered the stage, and with it several competing “open” standards, which (you’ve guessed it) don’t interwork. And management and orchestration (MANO) was introduced as an alternative sort-of NMS for the NFV world, of course alongside the above mentioned SDN crowd. And even the formerly well-organized and well-tranched operations support systems (OSS)/business support systems (BSS) layer is whirled around with lifecycle service orchestration (LSO) initiatives. The lowest common denominator appears to be YANG and (not YING) network configuration protocol (NETCONF). As long as a vendor shows those magic terms in the middle of his overloaded network-diagrams, operators have at least a glimmer of hope that some fine day someone will offer them a good solution for evolving their networks going forward. So what went wrong?

Well, a bit. The biggest mistake we as an industry made was starting the network-revolution at box level. Remember the structure of networks? The higher up you go, the more unified they become. That’s what Northbound interfaces are made for, and even 20 years ago, those were sort of standardized. The further down you go, the more proprietary all solutions become. All network vendors are using different languages for communicating with their boxes, and the way packet networks are managed can’t be applied to transport networks. Maybe someday, but not today. Something like simple network management protocol (SNMP) never made it to the optical camp, for the same reason why OpenFlow got stuck — physics are complex, and this was ignored completely.

So where’s the way out? We might have already found it, it’s just not as nice as some people want it to be. Packet networks were always more uniform and closer to the well standardized IT environment of data centers, and some of their nodes actually talk OpenFlow. Optical transport networks can be managed via domain controllers, and YING and YANG will do the rest. Are there real benefits? Yes, as soon as we manage to bring the right capabilities to the parent controllers, as soon as they will start talking to orchestrators that control data centers as well, innovative end-end services can be defined and turned up faster than ever, networks can be optimized in a way that was never there before, and automation will take control to improve operations. It just takes another few years, but how long did it take to get the first plain old telephony system (POTS) network up and running, after Mr. Bell had invented the telephone?