Network Resource Control for Grid Workflow Management Systems

Grid workflow management systems automate the orchestration of scientific applications with large computational and data processing needs, but lack control over network resources. Consequently, the management system cannot prevent multiple communication intensive applications to compete for network resources, which leads to unpredictable performance. Currently, the lack of control over network resources may prevent certain applications, i.e. applications that need high capacity and Quality of Service, to utilize Grids. Hence, such applications would use dedicated infrastructures. Because the costs to build dedicated infrastructures may far exceed the cost of using existing Grids, the Grid needs to support mechanisms to optimize the interworking between networks and applications. In this paper, we present the architecture and proof of concept to control network resources from Grid workflow management system and to manage network resources from workflow-enabled applications at run-time. Depending on the current network infrastructure capabilities or future advances, applications may employ existing QoS mechanisms or use application-specific ones to provide the desired network service. We believe that our approach leads to performance improvements in communication intensive applications and enables novel Grid applications, which require optimal interworking between networks and applications.