Cloud for Gaming
Cloud for Gaming refers to the use of cloud computing technologies to build large-scale gaming infrastructures, with the goal of improving scalability and responsiveness, improve the user’s experience, and enable new business models.
What Is Cloud Computing?
On-demand self-service: the ability to provide computing capabilities (e.g., CPU time, network storage) dynamically, as needed, without human intervention.
Broad network access: resources can be accessed through the network by client platforms using standard mechanisms and protocols.
Resource pooling: virtual and physical resources can be pooled and assigned dynamically to consumers, according to their demand, using a multitenant model.
Elasticity: from the customers’ point of view, the provider offers unlimited resources that can be purchased in any quantity at any time.
Measured service: cloud resource and service usages are optimized through a pay-per-use business model and are monitored, controlled, and reported transparently to both their customer and provider.
The typical interaction between cloud provider and customer works as follows: the customer connects to a “cloud marketplace” through a Web interface and selects the type and amount of the resources she needs (e.g., some virtual servers with given number of CPU cores, memory, and disk space). The resources are allocated from a large pool that is physically hosted on some big datacenter managed by the cloud provider. Once instantiated, the resources are accessed by the customer through the network. Additional resources can be acquired at a later time, e.g., to cope with an increase of the workload, and released when no longer needed. The customer pays a price that depends on the type and amount of resources requested (e.g., CPU core speed, memory size, disk space) and on the duration of their usage.
The deployment model defines the mode of operation of a cloud infrastructure; these are the private cloud, the community cloud, the public cloud, and the hybrid cloud models. A private cloud is operated exclusively for a customer organization; it is not necessarily managed by that organization. In the community cloud model, the infrastructure is shared by several organizations and supports a specific community with common concerns (e.g., security requirements, policy enforcement). In the public cloud model, the infrastructure is made available to the general public and is owned by an organization selling cloud services. Finally, the hybrid cloud model refers to cloud infrastructures constructed out of two or more private, public, or community clouds.
Cloud Computing for Gaming
The gaming industry embraced the cloud computing paradigm by implementing the Gaming as a Service (GaaS) model (Cai et al. 2014). Different instances of the GaaS paradigm have been proposed: remote rendering GaaS, local rendering GaaS, and cognitive resource allocation GaaS.
In the local rendering GaaS model, the video stream is encoded on the cloud as a sequence of high-level rendering instructions that are streamed to the player terminal (Fig. 2b); the terminal decodes and executes the instructions to draw each frame. Since encoding of each frame as a sequence of drawing instructions is often more space-efficient than compressing the resulting bitmap, the LR-GaaS model may require less network bandwidth than RR-GaaS and therefore eliminate the need for real-time video transmission capability. This comes at the cost of requiring a more powerful terminal with an adequate graphics subsystem.
Finally, in the cognitive resource allocation GaaS model, the game engine is logically partitioned into a set of modules that can be uploaded and executed at the client side (Fig. 2c). As the game evolves, the terminal receives and executes the appropriate modules and may keep or discard the unused ones. The CRA-GaaS model shifts the computation back to the client terminal, therefore reducing the load on the cloud. However, the client resources are used efficiently, since at any time only the needed components are stored locally. This is a significant advantage if we consider that the data of a complete modern game takes a lot of space for textures, 3D models, sounds, and code modules.
GaaS provides advantages for both game developers and players. The ability to offload some computation on the cloud allows simple terminals such as mobile devices to play complex games. Since the game engine is accessed on demand, flexible business models such as pay-per-play or monthly subscription can be easily implemented. Finally, game operators can scale up and down the amount of cloud resources used by the gaming infrastructure.
To effectively implement a cloud-based gaming infrastructure, it is necessary to address nontrivial issues related to game state partitioning, responsiveness, synchronization, and security.
Partitioning The key factor for achieving scalability of a GaaS infrastructure is the ability to partition the workload across the cloud resources. This is relatively easy if the workload consists of the execution of independent game instances that can be executed on any available resource, irrespective of where other instances are running. This is the case when the game does not allow different players to interact. Things become complex if the instances are not independent, as in the case of a MMOG system where all players interact with the same virtual world. In this case, the game engine must maintain a large shared state, allowing the players to “see” the effects of actions performed by the other players operating in the same virtual location.
This is achieved by partitioning the virtual world across multiple zones, each handled by a separate set of cloud resources. Given that communication between resource instances may incur significant delays, it is important that interaction across neighboring zones is minimized. For example, each partition may hold a collection of “islands” such that all interactions happen within the collection, while players can jump from one “island” to another.
Depending on the (virtual) mobility pattern of each player, some areas of the game field may become crowded, while others may become less populated. In order to cope with this variability, each zone controller is physically hosted on resources provided and operated by a cloud infrastructure. The cloud provider is in general a separate entity providing computational and storage resources to the game operator on a pay-as-you-go model. This means that the game operator can request additional servers and/or additional storage space at any time and release them when no longer needed. Thus, the game operator can request more resources when the workload on a zone increases, in order to keep the response time perceived by players below a predefined maximum threshold. When the workload decreases, the game operator can release surplus resources in order to reduce costs.
Synchronization The success of a gaming system is based on having players perceiving the game state as identical and simultaneously evolving on every player participating to a gaming session. If the game state is replicated in different cloud servers, a synchronization algorithm is needed to maintain the consistency of the redundant game state. To this aim, different schemes have been proposed in the literature (Game event synchronization 2006). They mainly differ from classic synchronization algorithms employed by distributed systems in their additional requirement for keeping the computation quick and responsive. To this aim, some schemes relax the requirements for full consistency during the game state computation.
A basic distinction is between conservative and optimistic synchronization. Conservative synchronization approaches allow the processing of game updates only when it is consistency-safe to do so. Lockstep (Fujimoto 1999), time-bucket synchronization (Fujimoto 1999), and interactivity restoring (Ferretti 2014) are some examples in the literature.
Optimistic synchronization mechanisms process game updates as soon as they receive them, thus increasing the responsiveness of the system. Yet, it is assumed that most updates are received in the correct order and that, in any case, it would be acceptable to recover later from possible inconsistencies.
Examples of optimistic approaches available in the scientific literature are the optimistic bucket synchronization (Diot and Gautier 1999), the combination of local lag and time warp proposed in Mauve et al. (2002), the trailing state synchronization (Cronin et al. 2002), and the improved time warp equipped with the dropping scheme and a correlation-based delivery control approach (Ferretti 2014).
Responsiveness The task of providing a pleasant experience to players becomes challenging when trying to deploy a large-scale and highly interactive online game. Responsiveness means having small delays between the generation of a game update at a given player and the time at which all other players perceive such update. How much such delays must be small depends on the type of online game. Obviously, the shorter the delay, the better. But it is possible to identify a game-specific responsiveness threshold T r that represents the maximum delay allowable before providing a game update to players. The typical T r for fast-paced games (e.g., first-person shooter, racing vehicles) is 150–200 ms, but this value can be increased to seconds in slow-paced games (e.g., strategic, role-playing games) (Ferretti 2014; Pantel and Wolf 2002).
A key point is that each player is geographically distributed. Thus, his latency to reach the game server on the cloud is usually different from other players. If a classic client-server approach is employed, it might thus happen that a responsive service is provided to some subset of users, while the other players can perceive a nonresponsive game evolution. This raises another main issue, i.e., fairness provision. This means guaranteeing that all players have the same chance of winning, regardless of their subjective network conditions (Ferretti 2014). To this aim, it should be guaranteed that all players perceive the same and simultaneous game evolution at the same time.
GaaS infrastructures represent an effective tool to provide responsive and fair gaming experiences. Cloud servers can manage the game state evolution in a scalable manner. Multiple server instances can be run in the same datacenter, when needed. Moreover, if the game involves worldwide distributed players, one might think to introduce a federation of cloud servers, geographically distributed, so that each client/player might connect to its nearest server. This could balance the network delays between the player and its server, thus augmenting the fairness level provided by the system. However, when multiple servers are involved, each one with a redundant copy of the game state, synchronization algorithm is needed to maintain game state consistency.
Security and reliability The security issues of GaaS infrastructures have become mainstream after the PlayStation Network outage that, in 2011, has halted the Sony online gaming network for 23 days. The network was shut down after detecting an external intrusion that led to a huge number of accounts being compromised and the exposure of the players’ personal information.
From the reliability point of view, large cloud systems provide some level of redundancy to cope with failures, including the use of geographically distributed datacenters, so that catastrophic events do not cause a complete outage. Unfortunately, the GaaS infrastructure may still represent a single point of failure; the PlayStation Network outage is just one example: in that case a security incident prompted the system administrators to temporarily shut down the whole service. Other possibilities must be considered as well: for example, the company operating the GaaS infrastructure may go bankrupt, depriving all players from the game service they might already have paid for.
From the security point of view, GaaS infrastructures are affected by the typical issues of cloud computing (e.g., insider attacks; Zissis and Lekkas 2012) and online gaming (e.g., cheating; Hu and Zambetta 2008). Online games are an appealing target for hacks because players often invest huge amount of time in their character development, and it is therefore quite easy to monetize game items on the black market. Additionally, individual accounts on online gaming platforms often contain information, such as credit card numbers, that are the typical target of cybercriminals. Details of the avatar of each player can provide information such as sexual preferences (Huh and Williams 2010) that could cause considerable embarrassment if made public.
- Cronin, E., Filstrup, B., Kurc, A.R., Jamin, S.: An efficient synchronization mechanism for mirrored game architectures. In: Proceedings of the 1st Workshop on Network and System Support for Games, NetGames’02, pp. 67–73. ACM, New York (2002)Google Scholar
- Ferretti, S.: Synchronization in Multiplayer Online Games, pp. 175–196. Wiley, New York (2014)Google Scholar
- Fujimoto, R.M.: Parallel and Distribution Simulation Systems, 1st edn. Wiley, New York (1999)Google Scholar
- Ferretti, S., Roccetti, M., Salomoni, P. Game event synchronization. In: Borko, F. (ed.) Encyclopedia of Multimedia, pp. 256–257. Springer, US, New York (2006)Google Scholar
- Huang, C.-Y., Chen, K.-T., Chen, D.-Y., Hsu, H.-J., Hsu, C.-H.: Gaminganywhere: the first open source cloud gaming system. ACM Trans. Multimed. Comput. Commun. 10(1s), 10:1–10:25 (2014)Google Scholar
- Marzolla, M., Ferretti, S., D’Angelo, G.: Dynamic resource provisioning for cloud-based gaming infrastructures. Comput. Entertain. 10(1), 4:1–4:20 (2012)Google Scholar
- Mauve, M., Fischer, S., Widmer, J.: A generic proxy system for networked computer games. In: Proceedings of the 1st Workshop on Network and System Support for Games, NetGames’02, pp. 25–28. ACM, New York (2002)Google Scholar
- Mell, P.. Grance, T.: The NIST Definition of Cloud Computing (Draft)–Recommendations of the National Institute of Standards and Technology. Special publication 800-145 (draft), Gaithersburg, Jan (2011)Google Scholar
- Pantel, L., Wolf, L.C.: On the impact of delay on real-time multiplayer games. In: Proceedings of the 12th International Workshop on Network and Operating Systems Support for Digital Audio and Video, NOSS-DAV’02, pp. 23–29. ACM, New York (2002)Google Scholar