When mobile data first emerged, much-discussed revolutionary business models failed to materialize. Bit rates for packet data technologies such as GPRS and EV-DO were moderate by today’s standards. High network latency prevented delay sensitive services such as conversational voice and interactive gaming, and devices were basic.
From a BSS viewpoint, data was generally treated as a ‘new product’ by the billing department. Charging for prepaid data was often based on ‘hot billing’ (nearly real-time) with charging systems typically dimensioned and licensed on the number of transactions per second.
BSS dimensioning reflected voice’s statistical predictability and a simple definition of transactions based on decades of experience coded into network planners’ DNA. Occasionally discussions would drift into reviewing the definition of a transaction, but these pretty much lapsed once the coffee had run out.
Fast forward 10 years - bits are tearing through the network at speeds approaching 100 Mbps, user plane latency is heading towards 10ms and devices are simply doing it for themselves. As a result, the humble transaction has witnessed something of a make-over. While voice transactions could be caricatured as impeccably behaved and deeply respectful of any network kind enough to accept them, data transactions are borderline psychotic and more than happy to wage war on any network or system that tries to control them. What is it that makes these data transactions so complex?
Essentially, it’s about complexity and variety. Let’s start with statistics. Voice was predictable and well behaved - average call duration, call arrival rate, permitted blocking - now where did that Erlang calculator go? Data statistics however are hugely variable and dependent on the mix of radio access technology, smartphone penetration and application mix - a mix that changes rapidly in time and geography. Measurements from a Tier 1 3G network reveal approximately 60% of sessions below 1 MB with only 4 to 5% of sessions accounting for nearly 50% of the total traffic. Session distribution over time is increasingly dependent on autonomous application behavior (synchronization, advert downloads etc.) so statistics are increasingly random.
Examples of such behaviors were nicely illustrated by the Sydney Morning Herald, late in 2012. They pointed to a free Sudoku app that drove requests for 10,000 ad images an hour for 36 hours. This resulted in 700 megabytes of data being consumed. On the weighty side, the Stagefright app downloaded 6.8 gigabytes of data over 52 hours, requested by just one customer. And this can be considered ‘normal’ behaviour. When an application server fails, the storm of transactions from the client apps trying to re-connect can push the needle off the scale.
Concurrency also drives data transaction complexity. A customer might check the weather, surf the internet and make a VoLTE call at the same time as their laptop is tethered to their handset for corporate VPN access. Perhaps the VoLTE call is being charged to a shared balance; maybe a family balance with a few members, or perhaps an enterprise balance with several thousand members. The associated load on real-time policy and charging systems can be huge and complex. And it doesn’t stop there…
Real-time ‘self-care’ interactions are an increasingly important feature of many services, for example notifications, balance queries and ad-hoc offers. Just because these transactions don’t enter through the network ‘front door,’ their load on any real-time system is no less significant.
In short, Data is Different. When you deploy that real-time voice BSS with 10kTPS stated capacity, you can expect 10kTPS. But when that same system is faced with data transactions, the effective rate can be reduced by as much as 90% due to the factors above. Beware - not all transactions are created equal!
Mobile data creates new and valuable opportunities for Communication Service Providers. But the associated network and charging transactions yield fundamentally new and complex challenges for real-time BSS systems, challenges that cannot be efficiently addressed simply by optimizing or scaling technology rooted in a much simpler voice world.