Whether in the role of a priest, death knight or wizard - in massively multiplayer online games (MMOGs), such as "World of Warcraft", millions of active users expect a disruption-free gaming experience. Moves should be displayed in real time, shared data and parallel play should be quickly networked and instant discussion should be made available via chat channels. This not only places high demands on PC performance but also on MMOG companies, explains Radu Prodan from the University of Innsbruck: "In order to ensure excellent gaming quality with a variable number of simultaneous players, MMOG companies require the availability of more server resources than they need on average to avoid quality suffering during peak hours." The result is an oversized infrastructure of thousands of computers, the ownership of which can cost up to 40 % of total game income. A more streamlined, flexible infrastructure would be possible by distributing the activities of MMOG companies using commercial cloud services. As part of an FWF project, the team headed by Prodan has now researched into how this role assignment could be used for the satisfaction of all sides.
Streamlining through sharing
As a basis for the model, an "ecosystem" for providing MMOGs based on a cloud infrastructure was developed: "Instead of combining all tasks under one roof, as is traditionally done at large MMOG companies, the required services are split into three smaller and better focused actors: game providers, game operators and (cloud) resource providers. The task of the game providers is to ensure good gaming quality in the face of fluctuating load conditions and differing resource availability," explains Prodan. In this model, game providers can rent and use the respectively required resources of several cloud providers flexibly, so that the service quality for end-users is consistently guaranteed. This is made possible by using in-house developed algorithms for load forecasting and smart load balancing.
Better reactions through simulation
In order to examine the practicality of this cloud-based hosting model as part of a performance test, a compatible simulator was developed which uses these algorithms. In addition, real world data gathered over the course of six months on 150 servers from RuneScape, one of the largest MMOGs on the market, was inputted into the model. Resource data from 16 currently operating cloud providers was also prepared for the simulator. One major challenge, Prodan adds, was making the services of cloud providers comparable. For the tests now published in the international magazine "Peer-to-Peer Networking and Applications", self-healing mechanisms in the case of technical disruption or sudden occurrences of resource problems were integrated. These help to keep possible gameplay disruptions as short as possible. The simulation runs showed that this self-regulating hosting with dynamic resource rental and flexible load balancing also led to gameplay disruptions of less than four minutes in the case of cloud resources with limited availability, irrespective of the duration of the underlying resource failure. Most of these faults affected less than two percent of game participants.
Virtual memory & real market fitness
The cloud-based hosting model developed as part of the FWF project allows a smart use of resources. This not only helps to protect the environment, but also allows lower increases in participation fees for end-customers on the back of the reduced infrastructure costs. The ability to forgo start-up investments in in-house infrastructure also allows smaller MMOG companies to participate in the market and break the existing monopoly. Smart load forecasting and load balancing mechanisms as well as self-healing mechanisms ensure a high-quality gaming experience, even in the context of the not altogether faultless conditions connected with using cloud resources. This also means that cloud solutions could be attractive for other non-game-orientated, performance-hungry IT applications.
Personal details Radu Prodan works at the Institute of Computer Science at the University of Innsbruck. For over 15 years, his research has focused on parallel and distributed computing. He earned his doctorate in 2004 at Vienna University of Technology and was appointed a professor at the University of Innsbruck in 2009. Prodan has contributed to numerous European projects and is currently the academic coordinator of the H2020 project ENTICE. Thanks to over 100 specialist publications, he held 11th place in the ranking of the most productive researchers worldwide in the area of cloud computing compiled by the magazine IEEE Transactions on Cloud Computing.
Publications and information