Anshul Rai, Ranjita Bhagwan, and Saikat Guha
Resource allocation is an integral, evolving part of many data center management problems such as virtual machine placement in data centers, network virtualization, and multi-path network routing. Since the problems are inherently NP-Hard, most existing systems use custom-designed heuristics to find a suitable solution. However, such heuristics are often rigid, making it difficult to extend them as requirements change.
In this paper, we present a novel approach to resource allocation that permits the problem specification to evolve with ease. We have built Wrasse, a generic and extensible tool that cloud environments can use to solve their specific allocation problem. Wrasse provides a simple yet expressive specification language that captures a wide range of resource allocation problems. At the back-end, it leverages the power of GPUs to provide solutions to the allocation problems in a fast and timely manner. We show the extensibility of Wrasse by expressing several allocation problems in its specification language. Our experiments show that Wrasse’s solution quality is as good as with heuristics, and sometimes even better, while maintaining good performance. In one case, Wrasse packed 71% more instances than a custom heuristic.
|Published in||Proceedings of the 3rd Symposium on Cloud Computing (SOCC)|
|Address||San Jose, CA|