Job schedulers for grids and clouds can offer great generality and configurability, but they typically do so at the cost of increased administrator complexity. In this paper, we present Wallaby, an open-source, scalable configuration service for compute resources managed by the Condor high-throughput computing system. Wallaby offers several notable advantages over similar systems: it lets administrators write declarative specifications of user-visible functionality on groups of nodes instead of low-level configuration file fragments; it presents a high-level semantic model of Condor features and their interactions and dependencies; it validates configurations before pushing them to nodes; it supports version control, “undo,” and configuration differencing; and it includes a networked API that enables extensions and advanced functionality. Wallaby allows administrators to extend pools to include more physical, virtual, or cloud nodes with minimal explicit configuration. Finally, it is scalable, supporting pools consisting of thousands of nodes with hundreds of configuration parameters each.
展开▼
