The Sustainability Impact of Scaling Strategies in Distributed Computing
Applications in the cloud can be scaled essentially either by “scaling up” (running on more powerful machines) or “scaling out” (running on more machines). The performance tradeoffs between these approaches have been studied extensively along traditional metrics. However, neither energy nor carbon has been studied so far. Given the macroscopic contribution of ICT (and in particular cloud computing) to global carbon emissions, this gap should be filled.
The primary goal of this project is to perform lab measurements of both traditional (time, memory usage, etc.) and sustainability (power) performance dimensions. Thereto, this evaluation will run well-known applications with existing implementations optimised for either scale-out or scale-up designs, respectively. The project will also compare different technologies of hardware components (CPUs, memory, etc.), depending on the availability of the hardware.
The project will leverage existing power measurement tools, both hardware- (external power meters) and software-based (internal power monitoring tools). As a by-product, the project results will provide a comparative analysis of the different tools, their ease of use, and their accuracy.
Milestones
- Get a first experimental setup running. For instance, a hash-join algorithm for databases on a baseline server and its scale-up and scale-out versions. Monitor power usage via EnergyAt/CarbonTracker and an external power meter.
- Extend the range of power monitoring tools (list to be discussed during the project).
- Run and test other distributed computing applications (list to be discussed during the project).
- Run and test the same applications on different hardware technologies (depending on hardware availability).
Supervisors
