ABSTRACT
The main contribution of this paper is to propose an improved schedule algorithm for the storm schedule system, which is used to replace the default Scheduler “EvenScheduler”. The default Scheduler will try to distribute executors and workers to all nodes via an even strategy. Since some components of them may require different kind of resources like Disk, Memory, CPU and Network, it may not be proper for all kinds of Topologies. In addition, in some specific scenarios, developer may hope the whole topology to running on a single supervisor node for reducing network flow. For other scenarios, when the topology is restarted because of upgrade maintenance or crash, developer may hope the corresponding components still running on the same nodes as before.
