Modeling the Scalability of Real-Time Online Interactive Applications on Clouds
Introduction
This paper is motivated by the popular and market-relevant class of Real-Time Online Interactive Applications (ROIA), such as massively multi-player online games, real-time training, e-learning based on simulations, etc. The users of ROIA usually pose very high requirements on the application’s QoE (Quality of Experience), including: short response times to user actions (about 0.1–1.5 s); frequent updates of the application state (up to 50 Hz); a very high number of simultaneous users in a single application instance (up to 104 simultaneously); and changing user numbers within an application session.
Recently, ROIA providers have started to use virtualized Cloud resources, e.g., via Amazon EC2 [1]. The high QoE demand requires application to be scalable, i.e., to be able to accommodate a changing number of users by dynamically adding/removing resources and employing load balancing for multiple servers in the Cloud. Although the Cloud resources (computational power and network bandwidth) are potentially unlimited, their economical use requires dynamically estimating both the effect and the overhead of load-balancing actions, i.e., adding resources or redistributing workload.
The new contributions of this paper are as follows: (1) we develop an analytical scalability model for ROIA which combines two parts - a model for computation resources and a model for communication (network) resources; (2) we extend the network model for a Cloud-relevant network topology with multiple data centers; (3) we implement and integrate the combined scalability model into our RTF-RMS – Real-Time Framework Resource Management System – for Clouds [2]; (4) we experimentally evaluate our model for a real-world use case of a shooter online game on the data center topology.
The next Section 2 describes the typical structure of ROIA and their execution in a Cloud. Section 3 presents our scalability model for predicting the maximum supported number of users per server. In Section 4, we illustrate how the proposed model is used to improve the load-balancing strategy in a resource management system. Section 5 reports the experimental results of our model for an example multi-player online game. Section 6 compares our approach to related work and concludes the paper.
Section snippets
ROIA: design and execution
In ROIA, users connect their personal computers (clients) to the servers and control their avatars that interact with application entities, i.e., other users’ avatars or computer-controlled NPCs (non-player characters). Each user can interact with other users within one virtual environment by sharing with them the common application state.
Fig. 1 shows the real-time loop model [3] for describing ROIA execution on hardware resources. Each client is connected to one server that processes users’
The scalability model
In this section, we first introduce the two parts of our model – computation and network – and then we explain how we combine them. We can model different instance types, usually provided by Clouds, by determining model parameters. A resource management system can then compare model predictions for different instance types as described in the next sections.
Computation model. We consider the two most important load-balancing actions w.r.t. the scalability of ROIA: (a) replication enactment adds
The combined model
In order to deal with both computation- and communication-intensive types of ROIA, we combine the predictions from our two models as follows:
The CPU model predicts the maximum number of users for a given number of replicas (denoted as), the maximum number of replicas () and the maximum number of migrations per second ( and). All these predictions are based on the tick duration. The CPU model assumes that all servers have the same computational power and, hence, each
Experimental evaluation
Fig. 6 shows a screenshot of our use case for experimental evaluation – the RTF-Demo application which is a fast-paced action game from the domain of FPS (First-Person Shooters), with all features of modern online games. In RTF-Demo, each user controls his own avatar (robot) in the 3D virtual world, and users interact by shooting at (and damaging) other users’ avatars. In our experiments, users are simulated by so-called bots which are dedicated client processes that autonomously trigger
Related work and conclusion
This paper develops a novel scalability model for Real-Time Online Interactive Applications (ROIA) that analyzes the application performance during runtime and predicts the demand for load-balancing actions on Clouds. In a practical case study of a multi-player online game, we demonstrated that our model is both adequate and accurate. Our experiments are conducted for the data center topology; we will experimentally evaluate our model for multiple data centers in the future work.
Previous work
Dominik Meiländer received his Ph.D. at the Department of Mathematics & Computer Science, University of Muenster. He works in the area of distributed and Cloud systems and has been actively participating in the EU Network of Excellence S-Cube, the EU “Marie Curie Actions” project MONICA and gained experience in cooperation with international industry within the EU FP6 project Edutain@Grid. Dr. Meiländer has published 17 papers (9 as first author) in reviewed international conferences and
References (21)
- Amazon Web Services, Inc., AWS - Game Development and Operations, 2016,...
- et al.
A dynamic resource management system for real-time online applications on clouds
- et al.
High-level development of multiserver online games
Int. J. Comput. Games Technol.
(2008) - et al.
Bandwidth estimation: Metrics, measurement techniques, and tools
IEEE Netw.
(2003) - et al.
A network-based scalability model for distributed real-time resource management
- et al.
Linear and nonlinear programming
- N. Jussien, G. Rochart, X. Lorca, Choco: an open source java constraint programming library, in: CPAIOR’08 Workshop on...
- et al.
Using a lifecycle model for developing and executing real-time online applications on clouds
- et al.
Linear Models: Least Squares and Alternatives
(1999) - et al.
Flying low: simple leases with workspace pilot
Lecture Notes in Comput. Sci.
(2008)
Cited by (0)
Dominik Meiländer received his Ph.D. at the Department of Mathematics & Computer Science, University of Muenster. He works in the area of distributed and Cloud systems and has been actively participating in the EU Network of Excellence S-Cube, the EU “Marie Curie Actions” project MONICA and gained experience in cooperation with international industry within the EU FP6 project Edutain@Grid. Dr. Meiländer has published 17 papers (9 as first author) in reviewed international conferences and journals.
Sergei Gorlatch is Full Professor at the Department of Mathematics & Computer Science, University of Muenster. Prof. Gorlatch has 25 years of international research experience in Computer Science and has published more than 200 reviewed papers and books. Prof. Gorlatch has actively participated in several national and international projects. He was an Executive Committee member & WP leader in the EC Network of Excellence CoreGRID and S-Cube (FP5), grant holder of the Tempus project ITSoftTeam and principal investigator in the EC projects Edutain@Grid and OFERTIE with international academic & industrial involvement. His special expertise lies in the fields of parallel and distributed algorithms, software and systems, as well as technology-driven e-Learning.