Leuven.
The next session here at EuroITV 2009 starts with another presenter from Alcatel-Lucent, Danny de Vleeschauwer. He notes the growth in catch-up television (CuTV), through initiatives such as the BBC's iPlayer or the ABC's iView - so that content is no longer consumed at exactly the same time (though still concentrated in a key period of time). This changes the requirements for IPTV: it can no longer operate under a broadcast or multicast model, but must now employ a unicast model which delivers a unique stream to each tuned-in viewer.
To maximise system performance, then, it is necessary to predict user behaviour - and a preliminary study has examined such patterns. It found that there are relatively common patterns in conventional TV viewing across the weekdays; similarly, there are common patterns in catch-up TV - the majority of views happen in the days immediately after first screening in the daily TV broadcast, progressively dropping off over the following days (and Danny presents a formula to describe this - key variables include the number of viewers who are interested in the first place, and the half-life of the specific content, which differs for example between news and drama shows). Interestingly, demand for some programmes drops off exponentially, for others following a power-law curve - and what determines which is which is an open question.
Such demand prediction can enable service providers to cache content in order to cater proactively for the likely demand. In essence, such a system would track the 'popularity' of specific television streams, and on that basis predict the further incidence of content requests - based on these predictions, the content held in the cache is chosen. Simulations of such access patterns can also be used to inform network infrastructure development: they enable engineers to develop a network which is likely to be overloaded only in extreme, unusual situations.
Further, on this basis, it is also possible to forecast network infrastructure requirements as the number of CuTV users increases, or as the amount of content hosted on the CuTV service each day increases. Notably, as is to be expected, caching has a substantial impact here, and it is also possible to model the impact of different cache sizes. Such caches ideally should be located relatively close to the end-user, incidentally.
