Design and Implementation of Packet-Aware Scheduling Algorithm for Serverless Architecture

Serverless is a fine-grained deployment model. Users only need to pay for the time segment of specific function execution to achieve true on-demand charging. However, the cloud function startup process requires downloading dependent packages, which delays the response time of the function and reduces the application's ability to respond quickly to load bursts. Therefore, a packet aware scheduling algorithm is proposed for the cloud function cold start problem. The algorithm establishes a scheduling queue and a packet cache tree at the working node. When the server receives the cloud function, the scheduler senses the optimal matching position of the packet through the improved location sensitive hash algorithm (WCFCM-LSH), and schedules the cloud function to the node queue; At the same time, in order to avoid excessive load on the work node, the scheduler will steal the tasks in the highest load scheduling queue and schedule them to other queues to reduce the load of the node. The experiment shows that the packet aware scheduling algorithm can effectively reduce the cold start time of cloud function, improve the scheduling efficiency, and reduce the waste of platform resources.