Since its beginning, the Internet as a system of systems has enjoyed unparalleled success as a powerful means of telecommunication. It is blazing fast, its infrastructure reaches even the most remote places on earth, and its design principles have shown to be quite robust. However, the opportunities that lie ahead of us will make greater demands on the Internet than being able to transfer emails or pictures. Already, well-connected societies have gotten used to video on-demand subscriptions, the streaming of major live events, and a ubiquitous IP capability of new devices. We may venture to project some trends:
- More connected things: More and more devices that have IP capabilities will find there ways into supply chains and homes. The current number of ~5bn units will grow four times to ~20bn units by 2020 (Gartner).
- More dynamic communication: The formerly unconnected-now-being-connected will more frequently communicate in real-time, which enables “dramatically faster cycle times” and “highly dynamic processes” (DHL, Cisco). In addition, various trends have pointed out that people are always on the move and need to be constantly in control of their physical devices while not being present at the source (Forbes).
- More traffic – delivered mostly by CDNs: IP traffic will grow at a compound annual growth rate (CAGR) of 23% from 2014 to 2019 (Cisco). An estimated 62% of all internet traffic will cross CDNs by 2019 globally, up from 39% in 2014 (Cisco). The 50% mark was surpassed in 2013 for internet consumer traffic in the US having been delivered by CDNs (Streamingmedia, Deepfield).
The last point highlights the importance of Content Delivery Networks (CDNs) in transporting content to the end user – in the right quality. The main idea behind this technique is to geographically distribute servers and then place content replicas on subsets of the server infrastructure (step 1: “generate abundance”). The second important step is then – whenever a user requests content—to select the “best” server to serve the content to the user. “Best” can relate to different parameters – a geographically close server may offer great latency, but can still be heavily utilized. Gathering the right information and making a well-informed decision is a CDN’s important step 2: “select a server”.
No serious web-based business today can do without some sort of CDN technique. This is reflected by the enormous share of global IP traffic crossing CDNs and the strong diversification that has occurred in the CDN space (Network World). Although, what the discussions around web performance techniques have largely neglected is the role of Internet Service Providers (ISP), who actually supply and manage the physical networks. Who would have thought that IP packets still have to pass the networks of the Verizons, AT&Ts, Telefónicas, British Telecoms, Oranges, Deutsche Telekoms, and China Telecoms (just to name a few) in this world. Every CDN (or any other performance improving technique) inevitably needs to interconnect with an ISP in order to be eventually networked to the end user. Furthermore, it is at this very frontier between the ISP and CDN universes where we actually observe a certain kind of idleness. An idleness, which, of course, breeds … opportunity.
Technical University of Berlin revealed that ~50% of server selections made by CDNs assign a suboptimal server to an end user (see the paper here). That means the network path between server and end user could either have been shorter, less utilized, more reliable, etc. The reason for this mismatch is obvious: CDNs just cannot see in real-time what the conditions at the edge or in the ISP network really are (not to mention what they will be). The two subsystems (CDN | ISP) are shielded from one another by design of current internet protocols. This does not mean that the information for alignment is non-existing. ISPs literally sit on it – the information is currently just not accessible or ready for use. This is a huge unexcavated potential which, if elevated, would significantly improve the quality of delivery and drastically reduce network Opex and Capex. The core idea is created from the observed “darkness” of ISP networks: shed light on network state information and provide it to all content delivery partners in the internet ecosystem, i.e. Raising All Visors!
BENOCS has implemented the idea and is already running its solution in a major European Tier 1 backbone network. Centrally collecting network inventory and performance information in real-time continuously feeds and updates our data engine. Exploiting NetFlow, IS-IS, OSPF, SNMP, DNS, among other protocols, provides the basis for our optimization and network state prediction algorithms. These “network state maps” are offered via standard API in a high frequency manner to CDNs, but are also branched off to internal units for Business Intelligence (BI). The benefits are clear: if a major ISP engages in this symbiotic communication with just its top 5 CDNs (by volume), peak link utilization in the network is sustainably lowered by ~30% (drastically driving down projected Capex figures for planned network capacity), and customer segments with mediocre delay numbers are almost entirely shifted into the sweet regions below 75 milliseconds.
Therefore, tear down the artificial walls for network-aware, smart content delivery: Raising All Visors!