Avaliação da técnica do Prebaking para redução do cold start em serviços FaaS para a runtime de Python.

Abstract

Platforms serverless are a fairly attractive business model for abstracting from the developer from the server infrastructure and allowing it to focus on the logic of your application. We adopt here the FaaS (functions as a service), which is a embodiment of serverless platforms where the client implements functions that are executed in the cloud in a stateless way. However, due to the collection model of these platforms, in which one pays only for the time when the suit function is executed, it is not advantageous for the provider which functions remain idle by occupying the cloud resources. That model brings with it the cold start problem: often, the time of instantiating a function can be very slow, which ends up warding the clients from FaaS. Prebaking’s technique, based on creating crafting functions of running functions and later restoring them later, in order to reduce the colld start, proved to be fairly effective for Java’s runtime. In this work, we find that for the Python runtime the improves is also significant: we were able to reduce the time of instantiation of a function by up to 1000%.