As a final part of my study of the Data & Knowledge Engineering master degree at Otto-von-Guericke-Universität Magdeburg, I defended my master’s thesis on 29.03.2017 under the title “Applying Placement Constraints on Server Consolidation for Enterprise Application Service Providers”. This thesis was accomplished as a part of the research done by the Fujitsu Lab at OVGU, and tried to solve an important challenge that faces the data center’s administration process.
The relevancy of the topic to the modern IT infrastructures comes from the increasing importance of both hosting cloud applications in remote servers, and storing more and more data every day. The emerging need of the data centers to perform those tasks rises a number of challenges such as the high energy consumption of those infrastructures.
This problem was solved by means of a virtualization technique called Server Consolidation, where the virtual machines (the services) are being packed in as few physical servers as possible. Finding such best packing can be very time consuming process and that makes approximate solutions acceptable ones. In addition to that, we cannot simply put any service on any server, because there are a number of critical conditions (constraints) that must be fulfilled by the final allocation. This exact point was the focus of the thesis.
The goal of the work was to identify the relevant types of those constraints and find a way to integrate them in the Server Consolidation process. In order to achieve the first goal, an intensive literature review about the related research was done, along with meeting with experts in the field of service providers. This approach ended with a list of 7 constraint types. Some of them handle the relationship between the service and the servers, like for example not allowing a specific service to be hosted on a specific server or visa versa. And some defined relationships between the services themselves, as being hosted on the same sub-group of servers, or being separated on different servers. Each constraint type aims at achieving one or more goal of performance, security, availability, compatibility, and saving administration cost and effort. The implementation of these constraint types within different algorithms was done while keeping in mind the need to provide a flexible and extensible design for future improvements.
After finishing the implementation, a number of experiments were used to evaluate it. The used evaluation method depended on a number of factors that measures the solution quality of the different algorithms on different data sets when applying various numbers and types of constraints. The results showed that Genetic Algorithms gives better solutions than the heuristic ones in terms of the number of violated constraints and the degree of such violations, but they clearly need more time to find such solutions. Another finding was that some constraint types are more likely to be violated than others, the thing that can be attributed to the relationships between services that such constraints need to hold.
This work showed that a number of different constraint types can be integrated into different Server Consolidation algorithms, and it also provided an evaluation framework for such integration.
We are very proud of Shadi and we are happy to obtain him as trained man for our team. You are interested in scientific research? Read on for more. Or you can call us at +49 4105 135 03 99, Email to sales@initos.com or via contact form.