PANACEA: Proactive autonomic management of cloud resources

Dimiter Avresky
Topics recommended for the 2016-2017 Work Programme: 

Panacea project creates Machine Learning Framework (MLF) for predicting anomalies in the cloud and a proactive autonomic management of cloud resources. In this research field, there are several issues that can be taken into consideration for future work programmes. PANACEA will allow users several advanced possibilities, based on the ML framework and the autonomic principles: - Proactive autonomic management of cloud resources. - Proactive software migration within one cloud. - Promising users anytime, anywhere access to their programs and data. - Developing new applications at the developer site. - Creating mission oriented distributed clouds with autonomic self* properties. - For owners of existing applications, with SLA requirements, efficient use of cloud resources will be ensured. - For owners of deployed applications to monitor their executions, by controlling and pro-actively managing them (VM migrations, proactive rejuvenation, predicting the time to crash). - Utilisation of Software Defined Networks (SDN) networks for implementing autonomic self* properties of Internet services and applications.

Projects major results: 

Up to now PANACEA generated the following outcomes: - Machine learning specification and training Version: v.1.0 for creating ML-based Prediction Framework for Internet services and applications anomalies. It is composed of several major steps: Initial Training Data Sets Collection, Lasso Regularization Process, Training Process, Prototype Implementation of the ML-based Prediction Framework, Impact of the Utilisation of ML Framework on applications availability, Availability enhancement by proactive machine rejuvenation. Fort coming steps: Resilient VM based topologies, dynamic reconfiguration, experimental environment for validation. - First prototype of experimental environment and software is implemented: TPC-W user, VM client (Apache Tomcat, TPC-W, MySQL, Feature Monitor client), VM Server (Feature Monitor Server, Online Monitor and Decision module, Machine Learning Framework). First experimental results are obtained: server response time as a function of the number of users from 16 to 128. The ML framework is validated in run time. The first prototype complete realisation is expected in the middle of 2015.

Potential exploitation strategy: 

Panacea Use cases selection is made based on several requirements, including impact on the society, its relevance and potential exploitation in the nearest future. Two Use cases have been accepted for implementation: 1) Cloud Web Hosting with a ML framework for a proactive cloud management With this use case, we could validate that we are able correctly to predict the remaining time to crash and response time of web servers, even when they are running in a quite dynamic environment and under heavy workloads. In addition, this use case will be invaluable in demonstrating that we can significantly increase the availability of web applications by means of software rejuvenation. 2) Second Use Case: Data Analytics as a service Data Analytics as a Service (DAaaS). The data analytics is based on Apache Hadoop7. It can use Map – Reduce algorithms written in Java. DAaaS represents an ideal application to be executed in a Cloud environment.

An update since the last Concertation meeting (March 2014): 

Discussions for preliminary Cross collaboration had been made during the Concentration Meeting (March 2014), Brussels with the following projects: - Portability and Interoperability in Clouds: contributions from the mOSAIC Project - - CACTOS – Context-aware cloud typology optimisation and simulation Project – Leader University of ULM. In addition, positive discussions for the future collaboration have been held with the leader of the project “Okeanos IAAS” - This is GRNET's cloud service, for the Greek Research and Academic Community.