The growing impact of video consumption on mobile device usage is clearly visible in the latest data on web activity conducted by CDN vendor Akamai. It also highlights how most mobile video consumption is over WiFi rather than cellular, with Apple’s iPad and Mobile Safari browser the clear leaders.
Akamai’s quarterly State of the Internet report has become the gold standard for global web statistics with increased interest in mobile, as the survey now partially separates WiFi from cellular usage, while also differentiating between 2G and 3G networks, although not yet LTE/4G. The report is based on data gathered from the Akamai Intelligent Platform, whose distributed architecture enables it to drill down in more detail than most other surveys, into regional and local trends, as well as separating out different traffic categories across both mobile and fixed services.
The report provides detailed analysis of attacks on web sites or whole sectors, and also now breaks down mobile traffic by browser type, but of greatest interest here is the information about video traffic that lies behind the headline data. The latest report for the third quarter of 2012 hardly mentions video, yet this is behind the most striking headline, a doubling in global mobile traffic over the year ending September 30 2012, although the pace appears to be easing, with the rise over the final quarter being 16%. It is wise to avoid placing too much store on an individual quarter’s results, as blips and statistical errors loom larger over a shorter time period, but it looks like the next year will see slower mobile traffic growth at around 70%, although this still represents a colossal gain from the current base. We can note in passing that on mobile, unlike fixed networks, traditional voice traffic is still growing at 6% per annum, although that figure too is declining.
For the first time Akamai analyzed browser usage across both PCs and mobile devices, with few great surprises but some interesting wrinkles that highlight emerging consumption patterns. One caveat is that currently the data is biased toward the US, since it was generated by sampling traffic across several hundred top-tier sites delivering content through Akamai, most of which serve predominantly US audiences. Akamai says it is working on expanding the data set to provide the same global view of browser usage as it has been doing for overall traffic patterns, but even now it does yield some wider insights on the relationship between video usage and service type, with naturally a huge difference between 2G and 3G.
The headline was that Android’s Webkit had retained its number one position on cellular networks, accounting for 37.6% of all requests, but that Apple’s Mobile Safari was closing in slightly on 35.7%. Opera Mini was number three at 19.3%, again with relatively little change over the year. Research In Motion’s Blackberry browser came in a poor fourth and actually losing ground at 4.2%, while Microsoft would not be thrilled by its 1.5% for its Mobile Explorer, even though this consolidated its fifth place. Around 10 additional browsers were also observed, only amassing 1.7% of usage between them.
The story changed significantly when WiFi access was added to the equation. Akamai did not unfortunately separate out the WiFi data, but a clear picture can be obtained by comparing overall usage with cellular-only. Microsoft will be more buoyed up by these figures, which revealed the impact of its rising tablet sales towards the very end of the quarter. Its overall usage figures had been hovering at just beneath 10% in third spot, but started to climb steadily towards the end of the third quarter, according to Akamai. Apple’s Mobile Safari browser was the runaway leader with its request level fluctuating at around 60%, with Android languishing at 23%, and then Opera Mini coming in fourth behind Microsoft Mobile Explorer at 5%.
We can take a stab at WiFi-only usage by first observing that WiFi traffic and access requests exceeds cellular for data. Taking this into account we come to the conclusion that Mobile Safari accounts for 70% of mobile WiFi usage, Android about 16% and Microsoft 14%, although with signs of a swing towards the last two setting in at the end of the period. Apple’s dominant position for mobile data usage is under threat from both Android and Mobile Explorer, although will still account for well over half until the end of 2013 at least. Note that these figures on WiFi do not include PC access both at home and in hotspots, which of course is still huge, noting for example that 56% of all the BBC’s online viewing of the 2012 Olympics was on PCs.
Akamai’s Connected Platform was less able to separate out mobile traffic between 2G, 3G (and in future 4G) so here it deferred to Ericsson, which with a customer base spanning over 1000 networks in 180 countries was deemed best placed to collect and analyze world mobile traffic. Again the results are not particularly surprising but indicative of overall trends in both usage and mobile device ownership. In the emerging era of 4G, Ericsson estimates that still almost 40% of installed devices are capable only of accessing over CDMA/GSM 2G, which still has significantly greater coverage than WCDMA/HSPA 3G. In fact operators are still building out 3G at a faster rate than 4G/LTE, having just passed the milestone of reaching 50% of the world’s population, while virtually all new devices sold now are at least 3G capable.
All devices can fall back to 2G when 3G coverage is not available, and if this happens a lot this has a significant impact on video access. Ericsson has mapped 3G to 2G fallback by location and also device type, finding a correlation between all three. 3G coverage naturally tends to be better in urban areas, and so we are not surprised to learn that mobile users in large cities are much more likely to own high end smartphones, while in a given country, penetration of such premium devices is two to three times less in rural areas where 3G coverage holes mean that fallback to 2G is a regular occurrence. This partly reflects lower income levels in rural areas, but also the lesser utility of a 3G device when the experience is constantly being frustrated by dropping back to 2G, when watching streaming video becomes effectively impossible.
There is also strong correlation between 2G traffic percentage and device type, starting at 100% obviously for 2G-only handsets, with ever lower values through smartphones, tablets and then mobile PCs at the bottom. This largely reflects a decreasing probability of encountering 3G coverage holes with lower levels of mobility. The more mobile a user is, the more likely that the service will fall back to 2G. There is a similar curve linking ratio of 2G to 3G traffic and operating system type, which in this case reflects the higher penetration of premium phones and tablets in urban areas.
The radio access technology also naturally has a big impact on application usage, with video, as well as file sharing traffic, virtually disappearing when only 2G is available. Here the survey reveals other factors coming into play relating to the specific service, such as data plans, traffic management policies and availability of free or cheap online media content.
What was perhaps slightly surprising was the degree to which overall video usage by a given user is sensitive to slight variations in the extent of 2G fallback. Ericsson noted a significant decrease in video usage even when only a few percent of a user’s total active time was spent on 2G fallback. This emphasizes the importance of delivering consistent as well as good user experience, which is a function of coverage for mobile services. It may be that this same factor also blights 4G services in the event of fallback to 3G, although in that event it would be possible to maintain some level of service with adaptive streaming. Some coverage areas are more important than others, and this finding highlights the wisdom of building out along commuter and major transport routes, where regular mobile users are most likely to be.
The data on fixed line usage and traffic patterns was less interesting overall but did reveal it is wrong to assume that average broadband access speeds will increase continuously. Akamai’s data showed a drop in Q3 2012 both in average global connection speed, by 6.8% to 2.8 Mbps, and in average peak speed by 1.4% to 15.9 Mbps. Average peak speed may seem like a useless measure from the user’s perspective as being unrepresentative of their quality of service, but does give the best measure of Internet access capacity. The reason for the decline in average rates is that while ongoing build out of fiber is increasing the number of people with high speed access at speeds of 10 Mbps or above, at the same time the number of users with low speed access is also rising.
Even so, the long term trend in average connection speeds continues to be upwards, with global average connection speed gaining 11% over the whole year to September 30th 2012, with growth in seven of the top 10 countries, and 104 altogether. The US was notable for making the strongest gains on the back of aggressive fiber rollout by Verizon, AT&T and others, with average connection speed up 8.8% in the third quarter alone, to 7.2 Mbps.
Apart from detailed analysis of major security breaches, which does not concern us here, the other major aspect of the Akamai report was its focus on migration from IPv4 to IPv6 for Internet transmission, which is of significant interest for broadcasting and video distribution. Of course the headline interest is that the world is running out of IPv4 addresses, as highlighted by Akamai with its observation that RIPENCC (the European Regional Internet Registry or RIR responsible for giving out IP addresses there) reached its last ‘/8′ block of IPv4 addresses (equivalent to 16,777,216 IPv4 addresses) in September 2012. RIPENCC then issued a statement that it was now effectively rationing IPv4 addresses, underscoring the importance that’ all stakeholders deploy IPv6 on their networks to ensure the continuity of their online operations and the future growth of the Internet.’
While IPv6 was developed to overcome the impending address shortage caused by IPv4 having space for only about 4.3 billion addresses, it has during its evolution acquired various features deemed valuable or necessary for the future of the Internet, some of which are particularly relevant for real time applications including video. Among them is in-built end-to-end device security via the IPSec protocol, which facilitates end-to-end authentication and identification of IP devices and so is useful for mobile TV. There is other good news for mobile TV, with support for global IP roaming, enabling devices such as audio codecs to go anywhere in the world and be contactable via a single IP address.
Broadcasters and particularly OTT operators have also welcomed IPv6’s embedded QoS support. While IPv4 was confined to ‘best effort’ IP packet delivery across networks, and therefore needed QoE mechanisms bolting on, IPv6 has a traffic-class field within packet headers allowing operators to prioritize data packets. This makes it much easier to ensure that audio/video content is delivered in uninterrupted streams across networks that are shared by less critical data traffic such as email. Of course this advantage diminishes in the case of a network almost entirely dedicated to premium video, but then such a network will probably involve some form of CDN anyway and so not rely on best effort delivery.
The remaining significant advantage of IPv6 for broadcasters lies in multicasting, with changes relevant for large scale OTT services or CDNs. Under IPv4, specific multicast routers are required to send multicast IP packets over IP networks, which tends to inhibit scalability to large geographical areas, because it may involve traversing long distances and many router hops for a service to reach the relevant multicast router serving a particular destination.
IPv6 fixes this via support for anycasting, which is sometimes described as a ‘fourth way’ of routing packets. Firstly there is unicasting, which is one-to-one, then there is broadcasting, which is one ‘to-all, and thirdly multicasting, or one-to-many but not everybody. Anycasting is one-to-one of many, where an IP packet’s destination is defined as any one, but just one, of a group of hosts or servers. It enables multiple hosts or servers that each provide the same service to share a single IP address. Then users can be directed just to the host nearest them to access the service, rather than having to go to a host the other side of the world. In this context nearest may not mean the geographically closest server, but the one that is accessible via the smallest number of hops, although the two will often coincide.
Anycasting is designed for services where a given set of users wants to access applications or data replicated across more than one server. It is particularly relevant for large scale video distribution, above all where interactivity is required when latency is a key issue. It means that the same content can be stored several times on different servers around the world, as is already increasingly done, and then users are directed to the nearest one. As well as reducing latency, this supports a degree of load balancing, depending on the scale, and also on how users are distributed around the region of operation. Configuration is also simplified, because there is just one single IP address to manage for all content, avoiding the need to correlate multiple DNS servers distributed around the network, DNS being the acronym for the Internet’s hierarchical naming system based on underlying IP addresses.
This in turn brings scalability. Instead of having to pump up a given server in a densely populated location, another server can be dropped in nearby and configured with the same anycast IPv6 address. Then users can be routed to the second server, and this idea can be propagated across a global network, enabling rapid reconfiguration to new routes in the event of congestion occurring. Here IPv6 does not achieve anything that cannot be done with IPv4, but makes life much easier, and the deployment will almost certainly be less expensive, more robust, and faster to converge or reconfigure in real time as required.
The ongoing migration from IPv4 to IPv6, still gathering steam, has had a distorting effect on Internet usage statistics as collected by Akamai, Ericsson and others. We all know without having to read a report that Internet usage is still rising just about everywhere, and so would be surprised to hear from Akamai that in the world’s biggest Internet market, the US, the number of unique IP addresses has actually declined over the last two years, as it has also in South Korea and even more significantly in Japan. These are all mature Internet markets with high levels of per capita connectivity, but even so in reality the number of users is still rising. The explanation is that because IPv4 addresses are running out, carriers are resorting to large scale Network Address Translation (NAT) to service still growing populations of devices. This enables actual addresses to be reused behind the NAT gateways. But it is only a temporary solution and just not workable for the impending Internet of Things requiring full end to end connectivity between huge numbers of devices. IPv4 combined with NAT would not scale effectively to such large numbers of devices and would impose too much complexity and processing overhead.
Akamai’s overall IPv4 address count is rising still, growing by 2% over Q3 2012 and 11% over the preceding year to reach 683 million, but that is also misleading. Some individuals account for more than one IP address, while others have only private addresses behind NAT gateways and so share a single IPv4 address, accessing the web via a firewall or proxy server. The second effect more than outweighs the first, so that Akamai estimates there are now well over 1 billion web users and a far larger number of connected devices already about equal to the global population of 7 billion and set to reach 50 billion by 2020, according to Ericsson. The great majority of those will be IPv6 connected by then but there will still be a rump of IPv4 during a prolonged period of dual running.