Dynamic bandwidth scalability in large scale networked virtual environments

Abstract

Ever since network technologies and applications have been developed, bandwidth has been considered a scarce resource. Although an increase in absolute numbers can clearly be observed (comparable but not exactly equal in trend to Moore’s law for processing speed), the demands put on the network infrastructure by applications have also surged in recent years. Several factors contribute to this fact, including the increased use of multimedia data (most importantly audio and video) and highly interactive application types. The latter are the subject of the thesis at hand, more precisely the class of applications known as ‘Networked Virtual Environments’. Fluctuations in the amount of users and the data flows associated with each individual entity present in these virtual worlds necessitate a means for limiting the amount of data to be either sent or received by each participant. To be able to adjust these requirements and resources at run-time would clearly be very beneficial for the scalability of the entire system. This thesis provides an overview of the network flows associated with and the bandwidth requirements of various types of Networked Virtual Environments. Ever since NVEs have been studied as a topic in (academic) research, ways in which to reduce the data flow have been proposed. They can be classified into four high-level categories: compression, aggregation, space partitioning and prediction. Each of these is discussed in this thesis, supported by examples, details and a summary of possible drawbacks. By combining these elements and comparison, the latter two techniques were found to be the most promising for further study. NVE technology has widespread applications, but the most well-known examples can be found in so-called massive multiplayer games. At their core, these games use engines for network functionality. A number of these (commercial) software libraries and architectures are discussed and investigated, revealing a wide variety of techniques and (sometimes ad-hoc) solutions to known issues. Each of these engines has specific optimizations that are linked to the genre of game they are developed for, indicating the fact that several solutions will have to be combined in order to obtain a generic solution that works well under most conditions. The implementation part of this thesis encompasses a generic and extensible Area-of-Interest system (the AOI-API), which includes the ability to discriminate several Levels-of-Detail in information streams. By generalizing the software implementation, the range of applications that can be supported grows substantially and extends even beyond typical applications genres (i.e. MMORPG, MMOFPS, MMORTS,...). Furthermore and as a showcase example for its flexibility, the AOI-API is integrated into an existing architecture which is intended to support large-scale networked virtual environments - ALVIC-NG. By combining ideas and elements from the NIProxy system, which is a platform for dynamic bandwidth partitioning over various network streams, a very powerful system is obtained that can be applied under many circumstances. This fact is supported by extensive tests, classified into seven main experiments, each highlighting either the versatility, bandwidth shaping capabilities or large scale features of the proposed solution. Although additional work is needed to investigate the applicability in more extensive scenarios and to quantify the net gains in terms of bandwidth for more large-scale setups, it is clear from the test results that the functionality of the AOIAPI, coupled with the NIProxy principles and the integration into the ALVIC-NG architecture is a promising step towards a truly scalable solution for client/server based networked virtual environments.