CACTOS: Context-Aware Cloud Topology Optimisation and Simulation

Jörg Domaschka
Ulm University
Topics recommended for the 2016-2017 Work Programme: 

Cloud data centre resource optimisation is a difficult and important problem. The scale of large cloud data centres alone motivates the need for research into improving service performance and resource utilisation in these environments. From the experiences in the CACTOS project gained so far, we have identified the following areas that need further research: With respect to cloud simulation, more detailed yet manageable and meaningful infrastructure models for data centres are needed that take into account the complexity of the real-world data centres’ virtualisation infrastructure. From a modelling and operator point of view prediction capabilities for non-functional properties of complex distributed software systems are just as necessary as approaches for merging Cloud data centre operation and architecture. Finally, we claim that a much tighter integration of data centre facilities with virtualisation is required. For instance, it is currently unknown how network bandwidth could be assigned dynamically to virtual machines based on their dynamic needs.

Projects major results: 

The overarching goal of the CACTOS project is to provide autonomic and interactive tools to support and improve the efficiency of cloud data centre operations. The three CACTOS tools CactoOpt, CactoSim and CactoScale build the foundation to craft, evaluate, and improve data centre design, operation, and placement optimisation algorithms. CactoOpt is designed to facilitate development of advanced optimisation mechanisms capable of both resource-level scheduling optimisation as well as holistic data centre-level optimisation. CactoScale provides data filtering and correlation analysis tools which will run on vast volumes of data generated from large data sets from both physical nodes and virtual machines. CactoSim enables the evaluation of optimisation strategies at design time based on various load and data centre models. All tools are integrated in the CACTOS Cloud data centre optimisation and management toolkit and their interplay shall improve energy efficiency, overall decision making, and allow accurate but low effort blueprint-based analyses. The functionality will be validated in three real world use cases ranging from operating cloud platforms over an enterprise application to the needs of scientific computing. At the current stage after year 1, we have established a model comprising physical and virtual components, as well as load patterns. We further enabled that CactoOpt is fed by the correct monitoring data including CPU load, memory usage. First mining algorithms have been established to trace the I/O behaviour of applications. CactoSim has been integrated with the live data centre topology optimisation prototype CactoOpt enabling the evaluation of arbitrary runtime optimisations at design time. A first initial public release of CactoSim is due at the end of September 2014.

Potential exploitation strategy: 

CACTOS will engage a coherent and converged exploitation strategy at the heart of which is the concept of proving concrete impact from the project results. Critical to the strategy of providing clear impact from project activity and sustainability beyond the project lifetime is identifying the target markets for CACTOS and then generating a community with this market in mind. A primary goal then, is building a community around the CACTOS tools in order to maximise impact and minimise effort for each member. With that in mind, the consortium currently evaluates the parts of the code to release as open source software, producing a set of exploitable tools mapped to use case validation and potential target users. This eases the community building and spreading the knowledge about CACTOS with the potential target groups including Cloud operators, Cloud infrastructure providers, data centre operators, cloud middleware developers and any company and scientific institution running a reasonably sized private cloud. In the period immediately after the project’s completion, the results will be used for educational purposes by all academic partners, but also for commercial purposes: Flexiant is going to integrate exploitable entities in their commercial offerings. University of Ulm will immediately apply the results and insights from CACTOS to their campus-wide data centre to ease its operation and offer automatised support for its scientific customers. Several partners envision a close future cooperation with other partners. Finally, partners from both academic and commercial domain are considering potential spin offs.

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

A cooperation including research and tools exchange with the CloudScale project has been established. Further, we are investigating the re-use of PaaSage tools for multi-cloud and cross-cloud deployments.