Scheduling jobs in a cluster by constructing multiple subclusters based on entry and exit rules

A system for scheduling a job in a cluster which comprises multiple computing nodes, said system comprising:

• one or more processors;
• a computer-readable storage device communicably coupled to the one or more processors;
• a rule definer executable by the one or more processors from the computer-readable storage device, the rule definer configured to define rules for construction of virtual sub-clusters of the multiple computing nodes, the rules including entry rules and exit rules for one or more of the virtual sub-clusters, the rules configurable to specify a plurality of characteristics for including a node of the multiple computing nodes into a virtual sub-cluster, wherein the plurality of characteristics include at least one member of the group consisting of a number of processors, a memory allocation, and a priority;
• a virtual sub-cluster manager executable by the one or more processors from the computer-readable storage device, the virtual sub-cluster manager configured to construct the multiple computing nodes in the cluster into multiple virtual sub-clusters based on the rules;
• a dispatcher executable by the one or more processors from the computer-readable storage device, the dispatcher configured to dispatch a received job to a selected virtual sub-cluster;
• and
• a sub-scheduler executable by the one or more processors from the computer-readable storage device, the sub-scheduler configured to schedule at least one computing node for the received job in the selected virtual sub-cluster;
• wherein the virtual sub-cluster manager is further configured to reorganize a virtual sub-cluster in response to a state change of a computing node that belongs to the virtual sub-cluster, the reorganization including the virtual sub-cluster manager configured to:
• determine whether the computing node satisfies an exit rule of the virtual sub-cluster;
• and
• exclude the computing node from the virtual sub-cluster in response to a determination that the computing node satisfies the exit rule of the virtual sub-cluster.

A computer-readable storage device having stored thereon computer executable instructions for causing one or more processors to perform operations for scheduling a job in a cluster including multiple computing nodes, the operations comprising:

• constructing the multiple computing nodes in the cluster into multiple virtual sub-clusters based on rules defining construction of virtual sub-clusters of the multiple computing nodes, the rules including entry rules and exit rules for the multiple virtual sub-clusters, the rules configurable to specify a plurality of characteristics for including a node of the multiple computing nodes into a virtual sub-cluster, wherein the plurality of characteristics include at least one member of the group consisting of a number of processors, a memory allocation, and a priority;
• dispatching a received job to a selected virtual sub-cluster;
• scheduling at least one computing node for the received job in the selected virtual sub-cluster;
• and
• in response to a state change of a computing node, reorganizing the virtual sub-cluster to which the computing node belongs, the reorganizing including:
• determining whether the computing node satisfies an exit rule of the virtual sub-cluster, and
• excluding the computing node from the virtual sub-cluster in response to determining that the computing node satisfies the exit rule.