James Pearce, VP Technology, Argogroup; 14th October 2004
Over the last 5 years, Argogroup has provided tools and platforms to enable the optimization of the user experience of mobile data services. With our Monitor Master solution, we provide device-focussed testing and monitoring of wireless browsing, messaging, and streaming applications. With our Device Master platform, we provide a structured approach to allow the capturing, storage and publishing of device profile information.
At the "Mobile Web Initiative" Workshop, Argogroup wishes to discuss and present our experiences gained from profiling and testing mobile devices over the last 5 years, and specifically, our desire to assist in the development of standards around this area: so that handset manufacturers, network operators and content providers can trustfully use device data to improve the quality of their services - from a mobile user's point of view.
This paper presents an overview of these experiences and plans.
Since late 1999, Argogroup (and an acquired company, AnywhereYouGo.com Ltd) has been capturing "profile" data about mobile devices. This has been for the purpose of simulating of devices when testing the quality of mobile services, and for the analysis of the compatibility and interoperability of those services with those devices. The device profiling platform (which is used both by Argogroup and our customers for) is known as Device Master.
The profiling process involves taking a real device - connected over a real network - and understanding how it behaves when accessing a wide range of content. Within the Device Master platform, we host a set of reference service applications that are visible over the wireless medium, and the profiler uses the device to visit each of the parts of the application in sequence. Naturally the applications (known as "assault courses", containing atomic "hurdles" of content) are designed to model the areas of diversity known to be present in mobile devices' data service support: for example, we are able to exercise the behaviour of the device with respect to WAP1.x and WML, WAP2.0 and XHTML, SMS, EMS, MMS, video, DRM, J2ME, and Unicode character set support, amongst many other standards.
As of October 2004, we have 36 of these content-typed assault courses, containing 633 discrete hurdles. Each hurdle is designed to produce one or many recorded attributes about the profiled devices. For example, the simple hurdle "Do you see bold text on the screen" is used to generate a single device attribute "Bold Supported".
Other hurdles are more complex and generate multiple attributes about a device's behaviour (such as the hurdle which asks "which of the following Unicode characters have rendered correctly?" which generates attributes relating to the device's support for each character). Finally, some hurdles are dynamic in nature - in order to efficiently seek the attribute's value. For example, the "Maximum URL Length" hurdle will automatically increment the length of the URL in a link until the device is no longer able to follow it.
The profiling process uses a dual-medium approach. The profiler is viewing the hurdle content on the mobile device in their (metaphorical) left-hand, whilst at the same time using a PC-based web browser with their right - which is providing an interface to allow them to enter the results that they are observing on the device. The sessions of the two browsers are associated together on the server-side, which means that there is a very tight, and rigorous, link between the profiling process and the accuracy and structure of the profile - which is being stored electronically on a server-side database.
As of October 2004, 450 mobile devices have been profiled by Argogroup in this way. This profile data is currently made available to our Monitor Master customers who then use those profiles to perform simulation and testing of their services. Many of our customers also perform their own profiling (for example of their own, private mobile device variants) - by accessing the Device Master interfaces over the Internet.
The data is syndicated to remote installations of Monitor Master using a web-based update process via XML over HTTP.
It is also possible to export this data from the Device Master database in a variety of formats. We can naturally produce web-based reports of the devices' capabilities, but can also export profiles to formats such as RDF/XML (mapping relevant attributes to UAProf where possible).
The growth of our profile structure, or schema, has naturally followed the evolution of data service support in mobile devices. For example, early profiles focussed on the devices' support of WAP and SMS, whereas newer devices need to be profiled for support of DRM, push-to-talk, streaming video, and other such emergent technologies.
We have traditionally oriented our profiling towards the greatest areas of diversity in devices. This has been a natural approach given that the prime usage for this data has been to perform interoperability testing with services. One of the benefits of this philosophy, however, has been that the profile definition also helps to identify (to, say, a content provider) which limitations, levels of markup support, or other boundary conditions are most valuable to consider when developing services.
We do not, for example, currently record the battery life or weight of a handset (because they have little impact on data service behaviour), nor, in the case of a mobile phone, the device's ability to make voice calls, since this is generally a boolean characteristic that varies little. We can (and do) however, focus our attentions on attributes such as the support for each and every WML markup element, the image formats displayed in MMS picture messages, and the devices' abilities to display UTF-8 characters in SMS text messages. We feel this is the level of detail that genuinely makes a device profile valuable - whether it is for service design, development, or interoperability testing and monitoring activities.
We are undertaking two major areas of research in the area of device profiling and profile syndication. These would be topics that we would be happy to present and discuss at the workshop.
The first is in the area of building device taxonomies and hierarchies. For example, we have observed that certain device attributes' values are common to, say, devices produced by the same manufacturer. Other attributes are generally common to devices that share the same browsers. Other attributes are common to devices that share the same sound chips, operating systems, camera and screen components, and so on.
Naturally it would be beneficial to be able to identify shared components (whether hardware or software) between device models or sub-revisions. It is also valuable to be able to identify whether attributes are shared between families or genres of components. (Do, for example, all Nokia browsers share their markup-support behaviours? Or do we need to split them into Series 30, Series 40, Series 60 genres?)
Part of this exercise also includes the taxonomisation of the components that make up a mobile device - to a degree sufficient to make this profile analysis effective. But it also includes the development of a standard taxonomy of the device attributes themselves, and contextualising attributes in such a way that they map to likely contributing components. For example, it is likely that sound-related attributes should be grouped together because they are the attributes that share the likely impact that the device's sound subsystem will have on them).
Building these hierarchies, and analysing the attribute values within them, then allows a very efficient way to scale the profiling process. It would allow a system to proactively deduce the behaviours a device may have (by comparing it with similar models) before it needs to be rigorously profiled. Since there are so many variants and versions of mobile devices around the world, we believe a smart approach to profile deduction in this way is vital to the success of such initiatives.
The second area of research and investigation is around the decentralisation of the profiling and syndication process itself. We can imagine an environment where a large number of parties are able to create device profiles (whether they be handset vendors, network operators, content providers, or independent organisations), and make them available in an open and publicly-published manner.
For example, it might be possible to create an ecosystem where device profiles are shared in a brokered market place. Participants have the ability to create and sign their own profiles, but also to take in and use profiles produced by others.
This approach could attempt to create a "critical mass" of profile production that would make the usage of such data an easy, efficient, and trustful exercise. Rather than an environment where profiles are generated by one constituency alone (say, device manufacturers), participants have the ability to compare, browse, and review profiles in a collaborative manner - much like the open-source code development paradigm.
We appreciate that the value of this environment is greatly enhanced (if not solely enabled) by the use of a standard approach to structuring the profile schemas, to gathering of the data in a reproducible way, and to ensuring that the data is used in a reliable and effective way (in, for example dynamic data service device targeting). This is certainly an area of discussion we would like to promote at the workshop.
Argogroup would very much appreciate the opportunity to join the workshop and provide what we think is valuable input - tempered by the experience of having worked on many of these challenges over the course of the last five years. We are committed to assisting the development of open standards and to the contribution of our ideas and findings to that process.
We trust that this Position Paper has clearly explained what we feel we can offer, and that it is viewed favourably.
Argogroup specialises in optimizing user experience for mobile services, a discipline that focuses on the interaction between network, service and handset. The company's two software products - Monitor Master and Device Master - help handset manufacturers and mobile operators test, monitor and develop products and services so that users spend more time and money on them.
Monitor Master is the world's leading User Experience Optimization platform, in use in over 40 mobile operators and equipment manufacturers worldwide. Device Master automatically creates detailed handset profiles so that all device data is held in a central knowledge base.
Argogroup is backed by Apax Partners and 3i. Its headquarters are near Guildford, UK, with offices in Seattle, San Francisco, Paris, Milan, Hong Kong, and Stockholm. For more information, please visit: http://www.argogroup.com/