HotCloudPerf 2018
Cloud and datacenter operators offer progressively more sophisticated service level agreements to customers. The Quality-of-Service guarantees by these operators have started to entail non-functional requirements customers have regarding their applications. At the same time, expressing applications as workflows in datacenters is increasingly more common. Currently, non-functional requirements (NFRs) can only be defined on entire workflows and cannot be changed at runtime, possibly wasting valuable resources. To move towards modifiable NFRs at the task level, there is a need for a formalism capable of expressing this. Existing formalisms do not support this level of granularity or are restricted to a subset of NFRs. In this work, we investigate the current support for NFRs in existing formalisms. Using a library containing workflows with and without NFRs, we inspect the capability of existing formalisms to express these requirements. Additionally, we create and evaluate five metrics to qualitatively and quantitatively compare each formalism. Our main findings are that although current formalisms do not support arbitrary NFRs per-task, the Directed Acyclic Graphs (DAGs) formalism is the most suitable to extend.