Microsoft Explores ‘Job-Centric’ Provisioning of Cloud Services

In the paper titled “Bridging the Tenant-Provider Gap in Cloud Services”, Microsoft researchers found out that cloud customers can purchase resources based on the “job-centric” standard. This cloud model will have and additional abstraction layer to cloud computing wherein an interface can be provided for customers to cite cost and performance goals as an alternative to an interface which allow them to allocate resources directly.

The research, which was released at the ACM Symposium on Cloud Computing 2012, was done by Microsoft Research Laboratory in the United Kingdom. The researches included Ant Rowstron, Thomas Karagiannis, Paolo Costa, Hitesh Ballani, and Virajith Jalaparti. According to them, cloud consumers will find a job-centric interface convenient and easy because it takes out the translation burden of high level objectives to the related resource necessities and it is flexible enough for cloud service providers to assign the amount of resources necessary to a particular job.

According to the research, cloud computing subscribers are responsible for aligning high level time objectives to particular resources required for a specific job. Although this is currently done, the paper revealed that consumers are only able to work out the number and size of virtual machines and not the shared provider infrastructure. Therefore, cloud service providers will find the job-centric model more applicable as they can combine multiple resources to complete a client’s particular job.

The researchers created a cloud foundation for MapReduce jobs with the use of Hadoop to explain the job-centric model. The Bazaar, the cloud framework they designed, provides subscribers a way to designate MapReduce job restrictions and completes the combination of resources which can best complete the job. It is able to explain the users’ goals to an array of resource tuples in order to predict the best tuple to complete the required job based on the condition of the datacenter at that specific time and allow the cloud service provider to receive more requests thereby increasing its revenue.

The process of translation includes the analysis of job details in order to find out the resource tuples which can meet the subscriber’s objectives before selecting the best tuple for the job. The Bazaar model, as the researchers claimed, was limited but can be expanded to permit allocating idle resources in order to complete a customer’s job earlier. Through the Bazaar, the researchers were able to show that a cloud service provider can accept more requests (about 3-14%) which can increase the datacenter’s goodput by as much as 87%.

The Bazaar model also implies that the cloud provider can charge based on jobs which take into consideration the desired completion time as well as the amount of data processing the job entails. With the job-centric model, both subscribers and cloud service providers benefit. The customer can know the upfront fixed costs and the performance while the cloud vendors can improve goodput and eventually result to higher revenue.