ICON Challenge on Forecasting and Scheduling
We consider the following problem: You are running a cloud computing service, where customers contract to run computing services (tasks). Each task has a duration, an earliest start and latest end, and resource requirements for CPU, Memory and I/O attributes. The tasks can be scheduled on one of multiple servers, each server has a limited capacity for the CPU, memory and I/O attributes. Multiple tasks can run concurrently on the same machine if the total resource consumption for all attributes is below the respective capacity. All tasks must be scheduled within their release and due dates, these dates are set so that no task stretches over midnight between two days. Tasks can not be interrupted, once started, they must run for their given duration. If a machine is used by a task, it must be running at that time. In addition to the cost of running the allocated tasks, the machine consumes some idle power if it is on. Every time a machine is switched on or off, a start-up resp. shut-down cost must be paid. All machines are off at the beginning of the planning period, all machines must be off at the end of the planning period.
The price of electricity for the data centre is a real-time price, and varies throughout the day. The actual price is not known in advance, a forecast must be used to generate a schedule. The total cost of the schedule is determined after the fact by applying the actual price of electricity to the energy consumption in each time period. One forecast of the price is given by the organizers. However there may be a large discrepancy between the forecast and actual price, illustrated in the figure, offering the opportunity to generate better forecasts based on historical data for demand and prices, and previous forecast information. Note that a forecast with a low error is not automatically guaranteed to lead to a schedule with a low overall cost.
You use the provided price forecast to generate your best possible schedule. All tasks must be scheduled, the cost of the schedule is given by the actual price for electricity. The validity and cost of the schedule generated is tested by a solution checker. You generate your own price forecast, and use a provided scheduling system to generate a schedule. The cost of the solution will depend on the quality of your price forecast. You generate both the price forecast and a scheduling system. You provide a schedule, its cost will be determined by a solution checker.