Utility Driven Elastic Services
International audience To address the requirements of scalability it has become a common practice to deploy large scale services over infrastructures of non-dedicated servers, multiplexing instances of multiple services at a fine grained level. This tendency has recently been popularized thanks to t...
| Main Authors: | , |
|---|---|
| Other Authors: | , , , , |
| Format: | Conference Object |
| Language: | English |
| Published: |
HAL CCSD
2011
|
| Subjects: | |
| Online Access: | https://hal.inria.fr/hal-01583573 https://hal.inria.fr/hal-01583573/document https://hal.inria.fr/hal-01583573/file/978-3-642-21387-8_10_Chapter.pdf https://doi.org/10.1007/978-3-642-21387-8_10 |
| Summary: | International audience To address the requirements of scalability it has become a common practice to deploy large scale services over infrastructures of non-dedicated servers, multiplexing instances of multiple services at a fine grained level. This tendency has recently been popularized thanks to the utilization of virtualization technologies. As these infrastructures become more complex, large, heterogeneous ad distributed, a manual allocation of resources becomes unfeasible and some form of self-management is required. However, traditional closed loop control mechanisms seems unsuitable for this platforms.The main contribution of this paper is the proposal of an Elastic Utility Driven Overlay Network (eUDON) for dynamically scaling the number of instances of a service to ensure a target QoS objective in highly dynamic large-scale infrastructures of non-dedicated servers. This overlay combines an application provided utility function to express the service’s QoS, with an epidemic protocol for state information dissemination, and simple local decisions on each instance to adapt to changes in the execution conditions. These elements give the overlay robustness, flexibility, scalability and a low overhead.We show, by means of simulation experiments, that the proposed mechanisms can adapt to a diverse range of situations like flash crowds and massive failures, while maintaining the QoS objectives of the service. |
|---|