- Grid Computing
- Cloud Computing
- Context-aware Middleware
- Mobile Data Management
- Multimedia Systems and Applications
- Mobile Computing
High speed network and increasingly powerful desktop machines boosted the emergence of grid computing that was intended to facilitate computational-intensive and data-intensive applications. After more than ten years of research, people start trying to apply grid services into operational usage over the Internet. One perspective is to leverage grid resources to support advanced mobile applications, which are driven by rapid evolution in versatile handheld devices and wireless networks but still challenged by innate constraints of mobile environments. Major issues, such as: user mobility, device energy deficiency and dynamic network connectivity remain to be tackled to improve system performance and reliability. Our MAPGrid project is one of the research efforts that address how to provide effective QoS for mobile applications by leveraging grid resources.
We focus on devising efficient resource discovery algorithms and data placement strategies for providing mobile users with multimedia services by leveraging heterogeneous and intermittently available grid resources. The objective is to support diverse QoS requirements of mobile applications while improving overall system performance in terms of client admission ratio, grid throughput and grid utilization. For instance, by exploiting the knowledge of client mobility patterns, device energy profiles and grid resource availability, we determine localized computational and storage resources within the grid for enhancing overall user experiences for mobile applications. We have applied techniques from graph theory, neural nets, etc. to deal with quality-aware discovery of grid resources for QoS-based mobile applications (e.g. streaming multimedia). We have devised predictive data placement techniques to cache multimedia segments on grid machines for better system performance. We also proposed an integrated solution that adapts to dynamic changes in device energy consumption and unpredictable grid resource availability without compromising application QoS.
The problem of resource discovery can be descried as: given heterogeneous and intermittently available grid resources and given a mobile request that may specify QoS requirements with possibile mobility information, how to selects optimal nearby grid proxies to service this mobile request. Because grid/cloud machines are intermittently available, one proxy may not be available for the whole service period of a mobile request. As a result of user mobility and heterogeneity of proxy resources, a previously scheduled proxy may no longer be optimal for the rest service period. Therefore, it is very important to define a criteria that can be used to evaluate the optimality of a grid proxy for a service period of a mobile request. Additionally, once a request is scheduled, context information (e.g. device energy, proxy resource availability) may be changed; adaptive strategies should be devised to handle dynamic changes in the system.
Data placement invovles the following issues: which proxy should replicate which data object, how many replicas should be created for each data object, when to execute these replications and how long the replicas should be kept on the selected proxies. Data placement techniques developed in Content Distribution Network (CDN) and grid systems are not sufficient for addressing these issues in mobile grid environment. This is because the presented unique scheduling model of grid-based mobile applications, i.e. one mobile request may be serviced by multiple proxies. In order to reduce replication cost in terms of storage capacity and network bandwidth consumption, mobile data placement on intermittently available grid proxies should be addressed at finer granularity both spatially and temporally. Spatially means a requested data object can be partitioned into segments and replication decisions are made to different partitions. Temporally means different replicas will be kept on intermittently available grid proxies for varying periods of time. Although short term caching is better to address on-demand requests, setup and tear-down of online connections for caching can be time consuming and affect users’ QoS experiences. In addition, since most mobile data caching decisions are made in a greedy manner based on individual mobile user’s context information, caching based on individual user’s mobility is not sufficient to achieve global optimization.
Grid-based Mobile Applications, Systems, Tests and Validations
Please MAPGrid project homepage for details.