My most recent task has been architecting the Playtech mobile product, which currently encompasses 10 games running on over 600 devices in 8 languages, each heavily customised for multiple licensees. Furthermore, the system was designed to continue scaling across more than a hundred licensees, with dozens of games and languages, supporting all mass market devices, all managed by a small team of inter-disciplinary experts. This has given me some interesting insights into the problems of fragmentation which I would like to share.
The dominant UI model on the desktop has been the Windows, Icons, Menus and Pointers system that became popular when Apple copied it from Xerox, and has stayed in the mainstream since Microsoft copied it from Apple.
No such standardisation is present in the mobile handset world. This will remain the case for the foreseeable future, as the market segments - offering “Swiss Army Knife” phones for power users, as well as specialised phones for people who value music, games, photography or fashion most.
Some will have keyboards, some trackballs, some touchscreens; some will have high resolution displays, some low; some will be landscape oriented, some portrait; some will have powerful processors and lots of memory, some won’t; and all of this will change radically over time, with some users upgrading every six months and some every four years.
Factor in the tight commercial deadlines that phone releases are subjected to, which never leave enough time for firmware QA, and you see why any mobile development platform that is large enough to be mainstream will be fragmented for a long time to come.
Anyone who has ever attempted to develop software for mobile phones knows that there are many hidden pitfalls along the way. However, mobile development need not be fraught with pain. A little experience immediately allows you to identify the key problems before you write a line of code, and design around them; a few rules of thumb can work wonders on development schedules and keep the QA team happy.
Pick Your Platform
I’ll try not to repeat my earlier post here, and will just briefly describe the three main types of mobile application platform – SMS, browser-based and installed application. Each has its own strengths and weaknesses, and they can often be used together to build a stronger product.
SMS is excellent for short, simple, immediate interactions across any handset - but break down once complex instructions are required.
Browser-based applications are growing hugely on the fixed web now that client-side scripting (such as AJAX) allows slick interaction, benefiting from automatic updates for every user – because everything is fetched from the server on-demand. AJAX on mobile is more theory than reality right now however, able to reach a tiny fraction of the market with incompatible implementations. Browser rendering is also hit or miss, and data costs can rapidly rack up as XHTML markup is heavy and flat rate data rare. Data cost is still an issue, with prepay customers regularly overcharged; flat rate, when available, is often full of loopholes and obscure clauses - billing per Kb is still very much the norm outside of the US.
Browser fragmentation will get worse long before it gets better – for example, radically different hardware like the iPhone introduces special case UI and design rules, and then 3rd party browsers like Opera or Mozilla further multiply these special cases whenever they launch for that hardware.
The other option is an installed client-side application, offering immediate and offline access, far greater presentation opportunities, and the potential for reduced data costs – but it is harder to update over time, and is best suited to services which are intended to be used regularly.
The Traditional View of Fragmentation
The most common thing people will say when discussing mobile applications, particularly those which are installed on the client-side, and Java apps in particular, is that fragmentation is a huge problem. This is generally repeated as a fact even by those who have never coded Java, which is why it is so often misunderstood. Figures like the 25,000 builds Glu required for Transformers are repeated with disbelief (which I share); I’d love to know how many of those included the tiny changes that operators demand to feature content on their portals, but I’d wager it’s a huge factor because I know Glu aren’t incompetent. As a comparison, we normally run to between 30 and 60 builds per language for an app or game.
Some history may explain how this feeling became “common wisdom” in the industry. In the bad old early days of MIDP 1 (the first version of mobile Java), applications were quite restricted in what they could do. MIDP 1 did not define a way to play sounds, vibrate the handset, or provide a mechanism to allow the game to take over the entire screen, for example.
Most major handset manufacturers saw this as an opportunity to differentiate their products, and dived in with proprietary APIs to do lots of extra things which arguably the MIDP 1 spec should have included in the first place. These could all be abstracted away by developers behind a common internal API and a single build, but it would have required a little effort to set up and was hardly ideal. This led to much of the initial bad press for mobile Java, which is still heard to this day. Buggy handset firmware is of course the other huge factor, but it is hardly restricted to the development environments - whole handsets regularly suffer.
Sun (the makers of Java) eventually got their act together and started to introduce extended functionality through the JSR process: first MMAPI for sound, then WMA for text messaging, and then MIDP 2. Rapidly, the old manufacturer APIs dropped away; some can still be found, but they are really only objects of curiosity these days. Right now, MIDP 2 is available on over three quarters of actively used Java-enabled handsets, with most popular devices also featuring lots of the common extensions as well.
The next most popular platform, Flash Lite, is now available on commonly used handsets in versions 1.0, 1.1, 2.0, 2.1 and 3.0. All of these offer different capabilities and implementations run in one of two different ways: either embedded on a web page (like a traditional Flash movie) or running as a standalone app (like JavaME content), each of which brings critical differences to the experience. Japanese handsets (where Flash Lite is arguably most popular) tend to embed Flash in the browser, preventing the Up and Down keys from being used in the Flash content. The main reason Flash Lite has fewer variations than JavaME is that it offers less power and access to the device’s features – and it still has its fair share of the standard cross-device issues and bugs.
After Flash come native Symbian OS apps – offering plenty of APIs (between the three major and many minor versions of Nokia’s Series 60 and the keypad and multiple touchscreen variants of UIQ). Recent versions of Symbian have repeatedly broken binary compatibility and the range of APIs always expand alongside device capabilities. So it goes on.
It is inevitable that any new feature absorbed into a phone – from cameras and music playing to GPS location tracking, OpenGL 3D and operator micropayment integration, will require a dedicated API before developers can access them. Features are being absorbed into phones at a very rapid pace, so any living development platform, be it browser or fully fledged open OS, will have API growth over time. There is simply no avoiding it.
Loosely drafted specifications with optional features and optional supported file formats still require management and should be avoided, but this is not rocket science. If you know that some devices support WAV sound but others only handle AMR, it’s not a particularly difficult problem to set up builds for the former to get WAVs and builds for the latter to get AMRs. We now have to generate platform-specific builds for a single application – we have certainly lost the “Write Once, Run Anywhere” bonus we were promised with Java. However this is simply unavoidable in the mobile space whatever platform you use, once you move beyond text-only messaging.
For what it is worth, my personal feeling is that the core Java functionality is becoming more reliable over time - we haven't had anything as bad as a 7650 or a 6600 for some time, and even some of the more notorious platforms like JBed are getting more stable. The more obscure APIs have teething problems and no platform is bug free, but most everyday features can be used easily enough once you know the basic tricks.
Mine’s Bigger than Yours
I’ve just listed one reason we may wish to split a single application into multiple binary files: incompatible file formats between devices. There are many more, and most apply to any development platform you care to mention.
The single factor which leads to the most extra work on any mobile app is the size of the device’s screen. The smallest screen the Playtech mobile product supports is 96x65 pixels for Nokia’s still popular low-end devices; the current largest is 352x416, used on some of Nokia’s highest end handsets; I have no doubt we will expand to support VGA and larger screens soon enough. In between, there are currently eight other screen size and orientation groups, all requiring individually redesigned graphics, to cover a few dozen actual real screen resolutions. For example, we consider a 176x196 pixel usable screen area to be equivalent to 176x220, and design the graphics and code to be happy with any height between 196 and 220px.
Note that often, the phone will let applications take over most of the screen, but keeps a header and/or footer for showing battery and signal information, the time etc.
Mobile apps and games almost exclusively use bitmap graphics – which have a fixed size and resolution – and not scalable vectors, like a traditional Flash movie, that would be capable of adapting somewhat to different screens sizes from only one binary file. Even Adobe admits that Flash Lite will need different binaries for different devices, as whilst Flash Lite theoretically offers the same vector scaling as its big brother, in reality this is too slow on current devices for fast complex graphics. Experts recommend the use of bitmaps for any non-trivial application, just like Java.
Managing this screen size complexity is a challenge without huge graphics production timelines (and suicidal designers), but one that can be met with scripting and templates. In Playtech’s case games can feature amazing levels of variation where the graphics require less than a day to customise for a new licensee, ready to drop into an automated build.
Fragmentation for All!
Assuming that intelligent code, good tools/scripting and carefully planned graphics can cope with screen size variations and differences in processor speed, what else do we have to worry about? Principally, the very first point mentioned in this post – the UI model of the handset, and the input mechanisms provided to make use of it.
The iPhone was rightly lauded for its touch user interface, a radical departure from even the conventional stylus-powered PDA platforms like UIQ and Windows Mobile, and thumb-powered interfaces as featured on the LG Prada. Windows Mobile itself is now changing to a new model based on gestures, and Nokia’s Series 60 is getting touch screens as well. Great news for consumer choice, but a fragmentation headache for developers - each of these systems requires its own dedicated interface solution to best meet the user’s expectations.
Even on more conventional phones, there are plenty of other differences – to enter some text into a field, most provide simple numeric keypads, but some handsets provide full keyboards and some provide their own proprietary solutions, such as RIM’s Suretype system. Even with a simple numeric keypad, some devices have a dedicated Clear/Delete key and some require a soft key for this function. In theory, you could just create a Java text field and let it worry about this detail, but that would tie you down to very ugly screens with extremely limited control. Most quality applications will implement their own UI components with full branding, but this means extra effort to support all types of character entry.
Even an app with no need for text entry should optimise the rest of the interface for whatever keys are available – allowing users to jump through lists by typing, for example, or handling motion control in games on devices which have no dedicated Up/Down/Left/Right keys. The optimal UI for a device with an Up/Down scrollwheel may be very poor on a device with a trackball, etc.
For a trivial app, these differences do not matter – and in a webapp, you simply can’t do these optimisations for your user experience. For a professional installed app that will see frequent and prolonged use, however, these optimisations are 100% critical to the project’s success. The interface must react exactly how the user expects it to, looking and feeling familiar whilst also presenting an attractive, professional and branded appearance.
Many years after Symbian was formed to be the phone OS of the future by some of the key mobile manufacturers of the day, the marketplace is still very fragmented. Motorola currently supports MIB, AJAR, MAGX (a Linux variant), Windows Mobile and UIQ (a Symbian variant); Samsung relies on a number of internal platforms alongside Windows and S60 (the other open Symbian variant); Nokia, the truly dominant force in mobile with between 30% and 40% of the global market at any given time, has managed to rationalise its OS portfolio down to the low-mid range Series 40 and the S60 platform for smartphones. Some operators are attempting to limit the number of platforms they support, but this is starting to look like trying to hold a finger in a leaking dyke.
If Nokia, by some metrics the most successful consumer electronics company the world has ever seen, finds it very hard to raise market share beyond 40%, then it would seem unlikely that anyone else can achieve more. Slow handset upgrade cycles mean that even if they did tomorrow, today’s fragmented handsets would still be in regular use in 2010. Current marketplace trends see increased diversification, both in form factor and in the types of software running on it – and with more software development platforms to work with. Fragmentation may be annoying, but it is here to stay.
Mobile development will always involve hitting multiple moving targets, and the only applications which succeed will be those that embrace this complexity and manage it sensibly. It is more important to ask “how many people can a platform reach?” than “how few targeted binary files will I have to generate?”
The only serious platforms that can give a satisfying answer to this question today are clearly WAP2 or Java, depending on the nature of the application. One day, it will hopefully become possible to write scriptable mobile web applications that can be updated day to day from the server end, but that simply is not possible for the mass market today, and won’t be for some years.
As CTO of a company which is in the business of knowing the ins and outs of every handset and every mobile platform, it is obviously to my advantage to state that only developers who can manage this complexity will be able to survive and thrive in the mobile space. Hopefully, the reasons why this is true, and the reasons why we as a company have taken the sometimes painful direction that we have, are clear. This is a very exciting space to be in, but when you try to work on the cutting edge you must take precautions not to get cut.