System Health Modeling for Improved Cloud Resource Provisioning

Abstract

Today's cloud systems comprise a large number of resources that are used for provisioning resource requests of multiple tenants. Each resource has its own set of configurable parameters that determine its overall health, viz., whether it would continue to function as designed. Any overloading of a resource in order to meet multiple tenancy requirements results in the system becoming "unhealthy", and therefore increasingly unfit to function normally. Typically, resources in a cloud environment are monitored using management tools such as IBM Systems Director and VMWare vSphere. While these tools do provide administrators with extensive information on the configurable parameters of monitored resources, they fail to provide a unified view on how "healthy" a resource is, especially when it is to be considered for provisioning a tenancy requirement. Such a lacuna usually forces systems administrators to make suboptimal guesses regarding the suitability of using the resource in question. To that end, in this paper, we discuss this crucial research issue and present our ongoing work on developing a provisioning health model for a resource, viz., its ability to meet a tenancy requirement taking into account its configurable parameters and current & past load conditions. We illustrate our model throughout our paper with a realistic running example, and discuss how our model could be developed and tested on reallife cloud environments.