Abstract
Pro-actively handling the fault in data center is a means to allocate the VM to Host before failures, so that SLA meets for the tasks running in the data center. Existing solution [1] on fault prediction in datacenter is based on a single parameter of temperature and the fault tolerance is implemented as a reactive solution in terms of VM replication. Different from these works, a proactive fault tolerance with fault prediction based on deep learning with multiple parameters is proposed in this work. In this work Cognitive Neural Network (CNN) is used to predict the failure of hosts and initiate migration or avoid allocation to the hosts which has high probability of failures. Hosts in the data center are scored on failure probability (FP-Score) based on parameters collected at various levels using CNN. VM placement and migration policies are fine-tuned using FP-Score to manage the failure proactively.
Highlights
Cloud data center has become a low cost solution for hosting of users computation and storage due to its unlimited resources on demand and pay as go model
Deep learning models have the capability to learn the high quality features and semantic relations automatically from the structured and unstructured information compared to previous machine learning models where features are selected manually
With cognitive neural learning model based on features extracted across all layers, the fault prediction capability is increased in the proposed solution
Summary
Cloud data center has become a low cost solution for hosting of users computation and storage due to its unlimited resources on demand and pay as go model. Even in case of host failures ensuring the continuity of user’s tasks with full or partial recovery in least latency is an important fault tolerance characteristic. Without it the quality of service of the data center is reduced and it will impact the business of data center provider. Proactive fault tolerance is a way to reduce the down time and ensure higher QOS for the data center. It involves prediction of VM or host failures in advance and corrective actions to avoid the failure or in case of failure, reduce the impact of failure. Dynamic replication strategy is proposed for VM in certain cases to reduce the impact of failure
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have