A methodology and a platform for monitoring multi-Cloud applications

Owing to the rapid progress in the Information and Communications Technology (ICT) and the subsequent demand for computing power, Cloud Computing, has received increasing attentions as an efficient method for handling workloads of great diversity and massive scale. It has been adopted in diverse areas, such as e-commerce, retail industry and academic environment due its various advantages including no up-front investments, lower operating costs, high scalability and elasticity, easy access and reduction of business risks and maintenance expenses. However, despite the undeniable advantages and considerable spread of Cloud Computing, there are still many related challenges and issues, which should be dealt with in order to ensure the efficient and proper operation of this new computing paradigm. Among the others, in order to ensure the quality of service and health of the running applications of the Cloud, monitoring the Cloud infrastructures and applications is a crucial task. In this work we propose a model-driven approach for monitoring applications running on multiple Clouds. The solution extends Palladio component meta-models and is based on the definition of Data Collectors, that are in charge of gathering data from resources (hypervisors, virtual machines, cloud services, application components), and Data Analyzers that receive data from Data Collectors and analyze them based on what is specified in monitoring rules. Monitoring rules can be defined by the system administrator or they can be automatically derived from QoS constraints defined during the design of the application. Data Analyzers are implemented by using C-SPARQL (continues-SPARQL) that is a query engine for RDF data, while Data Collectors are implemented using various approaches. One of these is Aspect Oriented Programming. This approach offers programming level mechanisms to define monitoring features separately from the application logic, and runtime support to weave the monitoring aspects into the execution of the application code. However, despite the undeniable advantages and considerable spread of Cloud Computing, there are still many related challenges and issues which should be dealt with in order to ensure the efficient and proper operation of this new computing paradigm. Specifically in order to ensure the quality of service and health of the running applications of the Cloud, monitoring the Cloud infrastructures and applications is a crucial task. The solution that has been proposed is defining Data Collectors that are gathered data from resources and produce as the output RDF triples data, and the Data Analyzer that at a high-level could be seen as accepting in input RDF stream data, and producing one or more streams of data in RDF format as the output. In this thesis the DA is implemented by using C-SPARQL (continues-SPARQL) that is a query engine for RDF data. In the present thesis, in the context of MODAClouds project, a monitoring framework will be developed which provides the enabling mechanism to trigger runtime adaption and to fill the gap between design-time modules and run-time application behavior. Aspect Oriented Programming, owing to its capability of keeping separate the cross-cutting features, has been chosen for developing the monitoring framework. Furthermore, meta-model for specifying the QoS, and meta-model and language for monitoring rules at design time. Finally, a methodology for the automatic/semi-automatic translation of QoS constraints into monitoring rules for translating them into queries is defined.

Data la rapida evoluzione dei sistemi di Information and Communications Technology (ICT) ed il conseguente aumento della richiesta di potenza computazionale, il Cloud Computing ha ricevuto grande attenzione dato che viene considerato una soluzione efficiente per gestire workload variabili ed eterogenei in grado di supportare richieste computazionali su larga scala. Il Cloud Computing è stato adottato in diverse aree, come e-commerce, sistemi per la vendita al dettaglio e sistemi informativi accademici dati i suoi svariati vantaggi che tra i quali i principali sono l’eliminazione di investimenti a priori, costi operativi inferiori, grande scalabilità ed elasticità, facilità di accesso, riduzione dei rischi e dei costi di gestione. Tuttavia, nonostante i vantaggi, esistono molte problematiche e sfide che devono essere affrontate per garantire un uso efficiente e corretto di questo nuovo paradigma computazionale. Tra queste, con l’obiettivo di garantire la qualità del servizio ed il corretto funzionamento delle applicazioni eseguite su Cloud, il monitoraggio delle infrastrutture Cloud e delle applicazioni è un aspetto cruciale. In questo lavoro di tesi viene proposto un approccio model-driven per il monitoraggio di applicazioni che sono eseguite su più Cloud. La soluzione estende il Palladio component meta-model ed è basata sulla definizione di Data Collector, che hanno l’obiettivo di raccogliere i dati dalle risorse (hypervisor, virtual machine, servizi cloud, componenti applicativi) e Data Analyzers che ricevono i dati dai Data Collectors e li analizzano in base a quanto viene specificato in regole di monitoraggio. Le regole di monitoraggio possono essere definite dall’amministratore di sistema oppure possono essere ottenute automaticamente dalla specifica dei vincoli di qualità del servizio definiti durante le fasi di progettazione dell’applicazione. I Data Analyzer sono implementati attraverso C-SPARQL (continues-SPARQL), un motore di interrogazione per dati in formato RDF, mentre i Data Collectors sono implementati attraverso vari approcci. Tra questi, è stato adottato anche l’Aspect Oriented Programming che offre dei meccanismi a livello di programmazione per definire le funzionalità di monitoraggio separatamente dalla logica applicativa ed un ambiente di supporto runtime capace di iniettare l’esecuzione del monitoraggio durante l’esecuzione del codice applicativo.