METHODS AND ARCHITECTURE FOR LOAD-CORRECTING REQUESTS FOR SERVERLESS FUNCTIONS

摘要

Methods and architecture for load-correcting requests for serverless functions to reduce latency of serverless computing are provided. An example technique exploits knowledge that a given server node does not have a serverless function ready to run or is overloaded. Without further processing overhead or communication, the server node shifts the request to a predetermined alternate node without assessing a current state of the alternate node, an efficient decision based on probability that a higher chance of fulfillment exists at the alternate node than at the current server, even with no knowledge of the alternate node. In an implementation, the server node refers the request but also warms up the requested serverless function, due to likelihood of repeated requests or in case the request is directed back. An example device has a front-end redirecting server and a backend serverless system in a single component.