Customer-oriented diagnosis of memory provisioning for IaaS clouds

Abstract

Infrastructure-as-a-service clouds enable customers to use computing resources in a flexible manner to satisfy their needs, and pay only for the allocated resources. One challenge for IaaS customers is the correct provisioning of their resources. Many users end up underprovisioning, hurting application performance, or overprovisioning, paying for resources that are not really necessary. Memory is an essential resource for any computing system, and is frequently a nperformance-limiting factor in cloud environments. In this work, we propose a model that enables cloud customers to determine whether the memory allocated to their virtual machines is correctly provisioned, underprovisioned, or overprovisioned. The model uses two metrics collected inside a VM, resident and committed memory, and defines thresholds for these metrics that characterize each provisioning level. Experimental results with Linux guests on Xen, running four benchmarks with different workloads and varying memory capacity, show that the model was able to accurately diagnose memory provisioning in 98% of the scenarios evaluated.