A unified framework for metering cloud environments

Cloud computing utilizes the integration of different computing technologies to achieve a utility computing model. Computing resources are consolidated and shared among different applications transparently. Cloud environments are like a market place, and an accurate cloud metering framework is needed for fair charge back, and accurate responsive SLA policies. Moreover, a diversified set of applications can benefit and make use of cloud metering such as cloud resource planning and scheduling, workload prediction, security attacks detection through events correlation, building highly available self-healing computing environments that can avail adequate resources in disaster times, etc. A data modeling approach coupled with a scalable distributed architecture is adopted to build our proposed programmable unified cloud metering framework. The proposed framework is based on a Cloud Metering Markup Language (CMML) and Network Transport specifications. The framework adopts a scalable multi-tier architecture template that can adapt elastically based on the target cloud environment size, functionality, and workload. The concept of autonomous metering data is introduced through the introduction of the Cloud Metering Objects (CMO). CMOs provide the ability to couple metering data with metering operations to be performed on the data which offers a means of explaining and interpreting the metering data. A full prototype of the CMML interpreter is implemented, as well as a Distributed Proc Filesystem as a transparent network communication protocol encapsulated within filesystem I/O operations. A case study is presented to demonstrate how the framework behaves in a realistic online shop environment. A set of functional experiments were conducted on the case study demonstrating the multi-perspective me- tering capabilities at different levels of abstraction throughout the underlying deployment cloud environment, namely data center perspective, service provider perspective, and end-user perspective. The Analysis of Vari- ance (ANOVA) and Generalized Linear Models (GLM) statistical methods were used to evaluate the proposed framework prototype from the performance perspective through three main experiments. An ANOVA/GLM comparative study between the Distributed Proc Filesystem protocol and the network Transmission Control Protocol (TCP) was conducted with respect to the probe effect, and a considerable save up in CPU cycles was demonstrated by the Distributed Proc Filesystem protocol over TCP. A framework end-to-end factorial design experiment based on ANOVA/GLM was conducted to study the different factors affecting the me- tering framework behavior. A detailed analysis of the different runtime factors and their proportionality is presented to show the factors effect on different metering engines of the proposed metering framework. More- over, the end-to-end factorial experiment demonstrated the low probe effect and execution overhead of the metering framework engines deployed on the cloud resources; mainly collection engines and transport pro- tocols. Finally, an ANOVA/GLM experiment was carried out on the case study environment to show the minimal effect of inserting metering probes, especially the application layer probes, on the web transaction response time experienced by the online shop application members