Burst Computing model and its comparison with FaaS

Identifier:  TFM:1635

Handle: https://hdl.handle.net/20.500.11797/TFM1635

Authors:  Molina Giménez, Enrique

Abstract:
FaaS (Function-as-a-Service) has gained significant popularity over the last few years, becoming a cloud service used by a large number of users for various workloads. In a quick and general definition, FaaS allows us to execute code snippets (functions) in the cloud without worrying about the underlying infrastructure (system administration, resource provisioning...). However, up to the current moment, FaaS only allows the atomic activation (and execution) of functions, without including notions of parallel execution of function groups. The proposed new model ”Burst computing” will address current limitations of FaaS, allowing: (1) trip-wire mechanisms for instant activation to launch massive groups of functions with guaranteed parallelism, (2) mechanisms for workload and/or data partitioning using unique identifiers within the group, and (3) group communication services and data aggregation, transparently leveraging node locality.. This thesis will focus on analyzing and proposing solutions for the three aforementioned improvements, and implementing a usable solution for the first two on the popular open-source FaaS platform Apache OpenWhisk.