A Preliminary Review of Function as a Service Platform Running with AWS Spot Instances

• Luciana da Costa Marques USP
• Alfredo Goldman USP

Resumo

Cloud computing enabled users to easily implement what was previously a complex data center infrastructure, reducing its maintenance related costs. However, cloud costs can be challenging to predict with all its variety of resources and price schemes. Amazon Web Services offers the spot pricing mechanism for virtual machines (VM), in which an end user buys idle VM capacity for up to 90% cost reduction. But, these machines can be terminated by AWS at any given time, so not every application may be a good fit using them. We here explore the deployment of function as a service (FaaS) platforms in spots by using funcX, a distributed and high performance FaaS project. We chose MASA-OpenMP, a DNA sequence comparison project, to be executed in two different scenarios: one with in-memory execution in spot VMs, orchestrated with HADS, a Hibernation-Aware Dynamic Scheduler for spot instances, and the other using a funcX client deployed in spot instances as well. Our initial results showed that costs using the funcX approach are potentially smaller, and it also offers more implementation flexibility. We also present possible next steps for more experiments and investigation.

Palavras-chave: cloud computing, spot instances, preemptive virtual machines, function as a service