Automatic driving simulation task scheduling method and apparatus, device, and readable medium

1. An automatic driving simulation task scheduling method, comprising:

constructing a network flow block diagram based on a cluster rack, a quantity of nodes and a quantity of processes, and encoding information of the network flow block diagram;

setting information of tasks to be simulated, and performing simulation task flow minimum path calculation via the network flow block diagram; and

updating a scheduling system state in real time according to a fairness strategy and a preemptive scheduling strategy, and scheduling based on the scheduling system state.

2. The automatic driving simulation task scheduling method according to claim 1 , wherein the network flow block diagram is configured such that a flow in the network begins at a vertex representing a task, passes through a plurality of directed edges representing the flow and edge costs, and a plurality of intermediate nodes representing jobs or physical machines, and finally enters a sink.

3. The automatic driving simulation task scheduling method according to claim 1 , wherein the constructing a network flow block diagram based on a cluster rack, a quantity of nodes and a quantity of processes comprises:

setting a root process node, a plurality of work process nodes and an unscheduled task node for each job, setting a computer node for each computer, and setting a rack node for each rack; and

connecting the plurality of work process nodes and the unscheduled task node and the computer node and the rack node by directed edges.

4. The automatic driving simulation task scheduling method according to claim 1 , wherein the encoding the information of the network flow block diagram comprises:

encoding the cluster rack, a computer node and a process node in the network flow block diagram; and

encoding the cost and weight of the directed edges in the network flow block diagram.

5. The automatic driving simulation task scheduling method according to claim 2 , wherein the setting information of tasks to be simulated comprises:

allocating flow to the directed edges such that a sum of the flow input by the plurality of intermediate nodes is equal to a sum of the flow output by the plurality of intermediate nodes; and

setting a directed edge assignment cost weight according to scenarios, participants and sensor configurations of the automatic driving simulation task.

6. The automatic driving simulation task scheduling method according to claim 1 ,

wherein the performing simulation task flow minimum path calculation via the network flow block diagram comprises:

obtaining a minimum feasible flow solution based on a cost-scalable push-relabel algorithm, wherein a sum product of a flow cost of an output edge of the minimum feasible flow solution is minimum.

7. The automatic driving simulation task scheduling method according to claim 1 , wherein the updating a scheduling system state in real time according to a fairness strategy and a preemptive scheduling strategy, and scheduling based on the scheduling system state comprises:

making the high-priority tasks preferentially occupy resources according to the preemptive scheduling strategy; and

setting an instantaneous resource allocation for a same-level priority tasks without preempting the resources of the same-level priority tasks according to the fairness strategy.

8. A computer device, comprising:

at least one processor; and

a memory for storing computer instructions executable on the processor which when executed by the processor, causes the processor to:

construct a network flow block diagram based on a cluster rack, a quantity of nodes and a quantity of processes, and encode information of the network flow block diagram;

set information of tasks to be simulated, and perform simulation task flow minimum path calculation via the network flow block diagram; and

update a scheduling system state in real time according to a fairness strategy and a preemptive scheduling strategy, and schedule based on the scheduling system state.

9. A non-transitory computer-readable storage medium, storing a computer program which when executed by a processor, causes the processor to:

construct a network flow block diagram based on a cluster rack, a quantity of nodes and a quantity of processes, and encode information of the network flow block diagram;

set information of tasks to be simulated, and perform simulation task flow minimum path calculation via the network flow block diagram; and

update a scheduling system state in real time according to a fairness strategy and a preemptive scheduling strategy, and schedule based on the scheduling system state.

10. The automatic driving simulation task scheduling method according to claim 2 , wherein each of the plurality of directed edges in the network flow block diagram is designed with two variables including a capacity and an edge cost; and

the capacity is used to limit the flow of a simulation task flow through each edge for cost calculation, and the edge cost is used to constrain a simulation task.

11. The automatic driving simulation task scheduling method according to claim 2 , wherein the method further comprises:

on the condition that a task starts running on a computer, adding an additional termination cost to the task on its side to all nodes except a node representing the computer in a cluster, wherein the additional termination cost increases with time and represents wasted resources when a penalty is forced to kill or move.

12. The automatic driving simulation task scheduling method according to claim 2 , wherein an outgoing edge of each computer is provided with a unit capacity to allocate a maximum number of computers and a minimum number of the computers for running simulation tasks by controlling a capacity between an unscheduled task node and the sink.

13. The automatic driving simulation task scheduling method according to claim 6 , wherein the cost-scalable push-relabel algorithm maintains a deviation from a minimum cost flow, and uses a continuous approximation method to solve.

14. The automatic driving simulation task scheduling method according to claim 6 , wherein the cost-scalable push-relabel algorithm comprises pushing and relabeling;

the pushing is applied to an edge to press the pre-flow of a beginning of the edge into a sink, and the relabeling is applied to a point and sets a height of the point as a minimum value of heights of all adjacent points plus one.

15. The computer device according to claim 8 , wherein the network flow block diagram is configured such that a flow in the network begins at a vertex representing a task, passes through a plurality of directed edges representing the flow and edge costs, and a plurality of intermediate nodes representing jobs or physical machines, and finally enters a sink.

16. The computer device according to claim 8 , wherein the construct a network flow block diagram based on a cluster rack, a quantity of nodes and a quantity of processes comprises:

setting a root process node, a plurality of work process nodes and an unscheduled task node for each job, setting a computer node for each computer, and setting a rack node for each rack; and

connecting the plurality of work process nodes and the unscheduled task node and the computer node and the rack node by directed edges.

17. The computer device according to claim 8 , wherein the encode the information of the network flow block diagram comprises:

encoding the cluster rack, a computer node and a process node in the network flow block diagram; and

encoding the cost and weight of the directed edges in the network flow block diagram.

18. The non-transitory computer-readable storage medium according to claim 9 , wherein the network flow block diagram is configured such that a flow in the network begins at a vertex representing a task, passes through a plurality of directed edges representing the flow and edge costs, and a plurality of intermediate nodes representing jobs or physical machines, and finally enters a sink.

19. The non-transitory computer-readable storage medium according to claim 9 , wherein the construct a network flow block diagram based on a cluster rack, a quantity of nodes and a quantity of processes comprises:

setting a root process node, a plurality of work process nodes and an unscheduled task node for each job, setting a computer node for each computer, and setting a rack node for each rack; and

connecting the plurality of work process nodes and the unscheduled task node and the computer node and the rack node by directed edges.

20. The non-transitory computer-readable storage medium according to claim 9 , wherein the encode the information of the network flow block diagram comprises:

encoding the cluster rack, a computer node and a process node in the network flow block diagram; and

encoding the cost and weight of the directed edges in the network flow block diagram.