Modality Cost Analysis: A Methodology for Cost Effective Datacenter Capacity Planning in the Cloud

In resource provisioning for datacenters, an important
issue is how resources may be allocated to an application such
that the service level agreements (SLAs) are met. Resource
provisioning is usually guided by intuitive or heuristic
expectation of performance and existing user model. Provisioning
based on such methodology, however, usually leads to more
resources than are actually necessary. While such
overprovisioning may guarantee performance, this guarantee
may come at a very high cost. A quantitative performance
estimate may guide the provider in making informed decisions
about the right level of resources, so that acceptable service
performance may be provided in a cost-effective manner. A
quantitative estimate of application performance must consider
its workload characteristics. Due to the complex workload
characteristics of commercial software, estimation of its
performance and provisioning to optimize for cost is not
straightforward. In this work we looked at breaking the
application into isolated modalities (modality is a scenario in
which an application is used, for example, instant messaging, and
voice calls are two different modalities of a media application)
and measuring resource cost per modality as an effective
methodology to provision datacenters to optimize for
performance and minimize cost. When breaking the application
into modalities, resource cost is assessed in isolation. Results are
then aggregated to estimate the overall resource provisioning
requirements. A validation tool is used to simulate the load and
validate the assumptions. This was applied to a commercially
available solution and validated in a datacenter setting.