This research addresses the problem of provisioning infrastructure for large-scale distributed applications. The proposed approach, a tool called CProv, takes the characterizations of application demands and hardware capabilities as input to automatically select the most cost-effective storage and compute configurations that meet the SLO requirement of the application. The main objectives of this project are 1) characterizing application behavior, hardware, and performance requirements with sufficient accuracy and manageable complexity 2) solving the constraint optimization problem.
- Leverage expertise at NetApp to inform how the CProv tool should characterize its inputs—cluster building blocks, applications, and SLOs.
- CProv needs to predict both the architecture and scale of various storage and compute building blocks required to meet input SLOs. They plan to evaluate CProv’s ability to predict the right scale by comparing with other capacity planning tools.
- With CProv, we will study when inﬂection points occur in cluster architectures (e.g., scale out architectures aren’t always cost-effective for a given workload) so that administrators who provision clusters can appropriately plan for these inﬂections.