TL;DR
This paper introduces three novel publish/subscribe (pub/sub) network designs for massively multiplayer online games (MMOGs) that aim to improve scalability and reduce latency. By integrating interest management directly into the pub/sub system, these designs offer a more decentralized and efficient approach to handling game state updates and replica distribution compared to traditional client-server architectures.
What Problem Does This Paper Address?
Massively multiplayer online games (MMOGs) face significant challenges in managing the vast number of players and game objects, leading to scalability issues with traditional client-server architectures. Distributing game state and ensuring low-latency communication for millions of concurrent interactions is complex. Existing distributed architectures often rely on separate interest management modules, which can become bottlenecks. This paper addresses the need for a more scalable, decentralized, and efficient networking solution for MMOGs by exploring how publish/subscribe (pub/sub) middleware can be leveraged not only for message dissemination but also for interest management itself.
Methods
The authors propose three distinct pub/sub network engine designs:
- Object-Based Network Engine: This design uses a topic-based pub/sub system where each game object has its own dedicated topic. Interest management is handled by a separate server. Updates are published on object-specific topics, and replica transfers are managed via point-to-point messages.
- Tile-Based Network Engine: This approach partitions the game world into tiles and uses topics associated with these tiles. It integrates interest management directly into the pub/sub system, eliminating the need for a dedicated IM server. Updates are published on tile topics, and replica requests/transfers are also managed through tile-specific topics, enabling multicast.
- Area-Based Network Engine: This design leverages content-based pub/sub to define areas of interest (AoI) using rectangular subscriptions. Like the tile-based engine, it integrates interest management. Updates are published with spatial coordinates, and clients subscribe to regions of interest. This approach aims for more efficient subscription management by using fewer, larger subscriptions compared to the tile-based method.
All three designs utilize pub/sub for both point-to-point (emulated via node-specific topics) and multicast communication.
Datasets
Not discussed in the paper. The authors used a custom MMOG research framework called "Mammoth" for their experiments, which simulates a virtual world with avatars and game objects.
Key Insights and Ideas
- Pub/Sub as a Decentralized Interest Manager: The most significant insight is the successful integration of interest management (IM) directly into the pub/sub system. This eliminates the need for dedicated IM servers, reducing potential bottlenecks and leading to a more distributed architecture.
- Trade-offs in Pub/Sub Granularity: The paper highlights clear trade-offs between topic-based (object-specific) and content-based (area/tile-specific) pub/sub approaches. Finer-grained subscriptions (object-based) offer precise control but can lead to more messages, while coarser-grained subscriptions (tile/area-based) improve multicast efficiency but might require more complex subscription management or less precise IM.
- Multicast for Scalability: The tile-based and area-based engines demonstrate the power of multicast for scaling MMOGs. By publishing updates and replica requests to topics representing tiles or areas, a single publication can reach multiple interested clients, significantly reducing broker load compared to the object-based engine's point-to-point approach.
- Obstacle Awareness Matters: The tile-based engine, with its obstacle-aware tile partitioning, shows an advantage in scenarios with many obstacles (e.g., walls). It can reduce update message traffic by effectively segmenting player groups.
Key Results
- Scalability: All three proposed network engine designs demonstrate good scalability, supporting a significant number of clients. The tile-based and area-based engines generally outperform the object-based engine, especially with more brokers, due to their effective use of multicast. With 8 brokers, they can support up to 900 players, while the object-based engine handles around 750.
- Performance Bottleneck: The primary bottleneck identified is the network bandwidth on the brokers, particularly the outgoing traffic. Response times deteriorate significantly when broker output traffic exceeds approximately 16 MB/s.
- Message Efficiency: The tile-based and area-based engines require fewer messages for replica transfers and subscriptions compared to the object-based engine, especially in scenarios involving group movement or when many players are interested in the same region.
- Area-Based Subscription Efficiency: The area-based engine generally requires fewer overall subscriptions than the tile-based engine, and its subscription count is less dependent on the specific area of interest.
- Impact of Game Scenarios: Player flocking and group movement significantly increase update messages. Obstacles can reduce these messages in obstacle-aware engines. Faster avatars increase subscription and replica transfer rates across all engines.
- Tile Size Optimization: For the tile-based engine, there's an optimal tile size (around 256 tiles in their experiments) that balances the number of update messages and subscription messages to minimize overall bandwidth usage.
Limitations and Future Work
- Limitations:
- The object-based engine's reliance on point-to-point communication for updates and replica transfers limits its scalability compared to the other two designs.
- The area-based engine's rectangular AoI definition is not obstacle-aware, which can lead to unnecessary subscriptions and replica requests in worlds with many obstacles.
- The performance of the tile-based engine is sensitive to tile size, requiring careful tuning.
- Future Work:
- The authors plan to investigate the applicability of their network engines to other location-based applications, such as traffic control systems.
- They aim to improve the area-based engine's interest management by making it obstacle-aware, potentially by extending the underlying content-based pub/sub engine.
Summary updated