Information-centric networking (ICN) is a topic that has been much written about in academic papers over the last several years, but is much less present in mainstream media. (For example, the Wikipedia article on ICN is very brief, and a Google search for ICN yields far fewer results than a search for software-defined networking.) Yet as our devices get more connected in the Internet of Things, and with 5G due to come out in a couple more years, new networking strategies may be needed. It’s worth taking a look at where networking is going and whether ICN could have a place in that world. (Note, there are related networking ideas – such as content-centered networking and named data networking – that won’t be discussed in this article, but merit their place in a longer discussion.)
The Future of Networking
Thoughts about the future of networking could naturally span over several articles. For this one, a solid overview will suffice. Billions of new devices will be connected to the Internet by 2020, due to the rise of IoT. This increased demand combined with increased power consumption is driving up memory costs in IP routers. In addition to these challenges, devices are becoming more mobile which can lead to packet routing inefficiency, and it is desired to keep latency low. With continued growth in IP routing tables, this low latency is not going to be a given. (Yes, this is a blitz through upcoming large-scale networking issues, but bear with me.) These problems will be difficult to solve under the current client-server model, but information-centric network architectures may be able to help.
One of the big changes in ICN architecture is that the client-server model is replaced by a publish-subscribe model. The publish-subscribe pattern is not new, but implementing it on a global level goes against the status quo. In this pattern, a requester of some information would not signal a specific receiver (e.g. a router) but instead request specific content without knowing who might have it. Then they rely on their nearest neighbors to produce the desired content; if that cannot be done, those neighbors ask their neighbors until the content is found. This is, of course, an oversimplification, but it is generally more efficient. It also allows for less reliance on DNS, which is only going to get more strained as more devices get connected (particularly as adoption remains sluggish for IPv6). A specific part of this model involves “allowing routers to use bits in the packets that are not addresses”, according to TCP/IP co-inventor Vint Cerf. This allows network administrators to “configure a network or network of networks around something other than formal address structures.”
Is ICN the Answer?
This may sound all well and good theoretically, but is this ICN approach really going to be the Next Big Thing? It depends on who you ask, but the answer seems inconclusive so far. In theory, ICN should work very well with increased virtualization, a la SDN/NFV, and some sources think it is necessary to ensure network quality in the upcoming years (as demand grows). And the publish-subscribe network pattern does work well with IoT devices that will be using the same model. But there are doubts. A 2011 research paper on ICN noted that ICN might bring a better security model and protect against denial-of-service attacks, but also notes, “We have raised doubts about whether ICN designs would improve network performance.” In 2013, Cisco published a series of slides on ICN, with the tagline, “Will it blend?” and avoided a conclusive answer. In the slides, Cisco Fellow Dave Oran says that ICN has promise, but remarks repeatedly that it is too soon to pass judgment on the methodology. Of course, it is mid-2016 now, but enterprises would do well to consider both sides of the argument when considering their business’s place stance on buying or selling ICN-compatible products.