Campus Compute Cooperative (CCC): A Service Oriented Cloud Federation

Author:
Prodhan, Md Anindya, Computer Science - School of Engineering and Applied Science, University of Virginia
Advisor:
Grimshaw, Andrew, EN-Comp Science Dept, University of Virginia
Abstract:

With the ever growing demand for computational resources, universities struggle to provide the cyber-infrastructure (CI) required to satisfy the needs of their researchers both in terms of quantity and diversity. Purchasing resources for peak demand for all resource types is not always plausible, and more importantly not at all efficient. Renting capacity from the commercial clouds can be an alternative, however, commercial clouds expect to be paid and are often very expensive. In this research, we present the Campus Compute Cooperative (CCC), which provides an alternative to commercial providers that yields increased value to member institutions at reduced cost. In CCC, member institutions trade their resources with one another in a market based environment to meet both local peak demand as well as provide access to resource types not available on the local campus. CCC also supports multiple differentiated qualities of service for the users.

Contrary to prior market based grid approaches, we focused our study towards a production level deployment. As a prelude to production deployment in multiple campuses we analyzed the challenges and prospects of such a federation with simulation. In this research, we simulated the CCC in a multi-campus environment with real production data traces from multiple institutions and evaluated the performance of the CCC with the current non-federated alternatives. Our simulation included data traces from three (3) institutions over the last quarter of 2017. From the results, we observe that, all the participating institutions can benefit from both the differentiated QoS and federation features of CCC and achieve a better net value individually and combined. Our simulation shows that CCC increases the overall net value of the federation by $350,000 in a month. This research also provides some interesting insights about the data traces which would help other prospective institutions to decide weather they could benefit from joining the federation or not.

Degree:
PHD (Doctor of Philosophy)
Keywords:
Market Based Grid, Grid Economy , Distributed Computing, Cloud Computing, Resource Heterogeneity, Grid Federation
Language:
English
Rights:
All rights reserved (no additional license for public reuse)
Issued Date:
2019/04/16