In order to satisfy customers by meeting service level objectives, data center operators are constantly forced to overprovision server resources, creating cost and inefficiency that puts extra pressure on the bottom line.
Hao Che, an associate professor in the Computer Science and Engineering Department at the University of Texas at Arlington (UTA) has been awarded a grant valued at $799,950 over three years by the National Science Foundation to develop his model for guaranteed delivery of service-level objectives without that overprovisioning.
Che’s project, which he is working on with Department Chair Hong Jiang and Professor Jeff Lei, is to create a model which allows customer service-level objectives to be mapped onto precise resource requirements at the individual server level. The abstraction will allow the model to be data center platform-agnostic and could lead not only to guaranteed service-level packages built to the specific needs of customers without overprovisioning, but also to helping cloud consumers purchase the resources best suited for their needs.
Last year, researchers estimated that some $30 billion worth of servers, or 30 percent of those deployed worldwide sit “comatose,” which could indicate huge potential savings for data center providers.
“This model can also benefit cloud customers,” Che, a former systems architect, said in a statement. “Today’s cloud service providers mainly provide resource-centric services to their customers, offering hardware and software resources only, leaving the hard question to the customer themselves to deal with. For example, how many resources are needed to meet the service-level objectives. With our model, the customers will be able to know exactly what and how much resources need to be purchased to meet their needs. This approach should provide important insights into computer systems and architecture designs with an eye toward performance guarantees, cost-effectiveness and better utilization of resources.”
UTA has developed a strategic plan to emphasize data-driven discovery, and department faculty are working on several related projects.