Sudden crowds are part of the Internet: a blog post gets popular, and people flood in, sometimes straining servers to the breaking point. Online virtual worlds are subject to the same phenomenon, and unlike with a blog site, visitors to virtual worlds can't be spread across different servers arbitrarily, to balance the load. Friends need to be kept together, so that they can interact. Now VastPark, an Australian company that provides foundations for virtual worlds, is planning to use new technology from National ICT Australia (NICTA), a research institute, to solve this problem.

NICTA's system incorporates peer-to-peer networks, which help reduce the load of sudden crowds by getting bandwidth and processing resources from each new user who makes a demand on the network. Santosh Kulkarni, a senior researcher in the network information processing group at NICTA, says that the peer-to-peer networks will also reduce the cost of infrastructure for companies who use it in their virtual worlds, since the system allows more users to sign up for a world, without requiring the company to support them with more servers.

The typical network architecture for virtual worlds, Kulkarni says, involves central servers that control all the information flowing to and from the clients installed on users' computers. Some virtual-world architectures, such as that of Linden Lab's Second Life, stream all the information about the world from those central servers, including information about 3-D content, along with information about the position of the user's avatar. Other architectures, such as that of the Multiverse network, separate information about the look of the world from information about how avatars are interacting. Display information gets distributed with client software and stored on users' computers, reducing the amount of information that needs to pass through the central server on a regular basis. Kulkarni says that NICTA's system reduces the infrastructure required by the company hosting the world because, while the company can still run a central server that distributes client software and contains information about 3-D content, peer-to-peer networks can handle information about avatar position and character interaction.

Kulkarni says that figuring out how to map this content onto a peer-to-peer network is a completely different problem from figuring out how to divide content for common peer-to-peer applications, such as file sharing. NICTA's technology, he explains, divides the space in a virtual world into regions, and peers become responsible for hosting the regions. As the regions become more populated, they are further subdivided, and more peers become responsible for the pieces. When a user's client wants to find out about the objects around her avatar, it sends a request to the network, which finds the peer hosting her current region. That peer puts the user's computer in touch with the peers closest to her in the virtual world, who have information about her surroundings. The system also contains code to reduce the load on the network, such as an algorithm that notices clusters of users moving around the world together and consolidates their update requests into a single query.