Search

Optimisation of (the configuration and deployment of) distributed cloud applications is a complex problem that requires understanding factors such as infrastructure and application topologies, workload arrival and propagation patterns, and the predictability and variations of user behaviour. This chapter outlines the RECAP approach to application optimisation and presents its framework for joint modelling of applications, workloads, and the propagation of workloads in applications and networks. The interaction of the models and algorithms developed is described and presented along with the tools that build on them. Contributions in modelling, characterisation, and autoscaling of applications, as well as prediction and generation of workloads, are presented and discussed in the context of optimisation of distributed cloud applications operating in complex heterogeneous resource environments. TRUE pub

RECAP targets the automated operation and management of applications / service chains in largescale geographically distributed infrastructure. As such, RECAP components need to operate across distributed infrastructure, which makes the RECAP tooling itself a distributed application. Earlier documents introduced RECAP’s initial architecture (D4.2) and described the initial prototype of the platform (D4.6). This document combines both predecessor documents and enhances them with an updated description of the RECAP architecture. From an architectural point of view, RECAP consists of four functional building blocks (sub systems): (i) landscaping and monitoring, (ii) application and infrastructure optimization, (iii) simulation and planning, and (iv) data analytics and machine learning. (i) The Landscaping and Monitoring sub-system is responsible for gathering information about the structure of the current infrastructure and application landscape and monitoring their state. This is the main source for both the Data Analytics and Machine Learning as well as Application and Infrastructure Optimization sub-system. (ii) The Application and Infrastructure optimizers make decisions based on internal models as well as live information provided by the landscaping and monitoring subsystem. Separating between dedicated application and infrastructure optimizers enables the realisation of a ‘separation of concerns’. (iii) The simulation and planning sub-system provides means for supporting the validation of RECAP models, the execution of ‘what-if’ scenarios, and the long-term planning of the large scale infrastructures. (iv) The data analytics and machine learning sub system provides tools and means to distil statistical properties and patterns from load traces; particular focus is thereby put on workload prediction. With regard to integration, RECAP targets a loose coupling between the components fostering independent uptake and re-use of building blocks.

This document is accompanying material for the prototype deliverable D5.2.2. It describes the changes for the second version of the CACTOS Toolkit and provides details on the integration between the tools and toolkits. A main focus is on showing updated models, as this is how information is passed between the tools. Identical models are used during Runtime and Prediction time. Please refer to accompanying material for the prototype deliverable (D5.2.1 CACTOS Toolkit Version 1) for an overview on the CACTOS toolkits and an exemplary use case. Note that there are two CACTOS toolkits: The CACTOS Runtime Toolkit (label before year 1: CACTOS Toolkit) and the CACTOS Prediction Toolkit. The CACTOS Runtime Toolkit contains the tools CactoScale and CactoOpt and is described in this deliverable. The CACTOS Prediction Toolkit is described in (D6.4 CactoSim Simulation Framework Final Prototype). The major architectural additions to the CACTOS Runtime Toolkit since year 1 are added support for monitoring and scaling of White-Box Applications. White-Box Applications allow for monitoring of application internals on top of the VM-level metrics that CACTOS collects for all VMs. White-Box Applications such as PlayGen’s DataPlay can use CACTOS AutoScaling services to let the CACTOS Runtime Toolkit adapt the degree of horizontal scaling based on the current load. This document describes both the additions to the models and CACTOS Runtime Toolkit that have been made to support monitoring and scaling of White-Box Applications. Finally, the document provides detailed insight into the architecture and service structure of the CACTOS Runtime Toolkit. This includes a detailed description of the Virtualisation Middleware Integration (VMI) and VMI Controller that form the Cloud-specific connector CACTOS uses to translate its optimisation decisions to a running Cloud environment’s API. Additionally, an overview over the Extensible Services Infrastructure architecture style is given. This architecture style allows for a dynamic reconfiguration of used optimisation algorithms and policies. The style also eases the coupling and analysis of optimisation algorithms in the CACTOS Prediction Toolkit.

The CACTOS project aims to improve the operational efficiency of cloud data centres by supporting data centre operators in the planning and operation of heterogeneous data centres. One major goal of CACTOS is to enable automated capacity and resource management for virtualised infrastructure environments built upon the Infrastructure as a Service (IaaS) paradigm. This document outlines the model‐driven methodology developed for the integration of runtime monitoring of cloud‐based data centres with runtime optimisation techniques. The CACTOS project develops an integrated solution for runtime monitoring, optimisation and predictive analysis of data centres. The solution supports data centre providers in managing and planning data centres. CACTOS consists of two toolkits: • The CACTOS Runtime Toolkit enables automated resource planning and optimisation for IaaS data centres. • The CACTOS Prediction Toolkit supports what‐if analyses for existing or planned data centre topologies that account for effects caused by automated resource optimisation. While the focus of this document is to describe the integration methodology developed to couple runtime monitoring and optimisation for cloud data centres in the CACTOS Runtime Toolkit, the outlined methodology was developed to facilitate the integration across all toolkits developed in CACTOS. Hence, the integration of optimisation and monitoring with the simulative predictions in the CACTOS Prediction Toolkit is also discussed. The main contributions of this deliverable are the CACTOS Cloud Infrastructure Models that define the common language through which the runtime analytics tool, CactoScale, and the optimisation tool, CactoOpt, exchange information on the data centre’s structure and operational state. The models allow for the capturing of the deployment of Virtual Machines (VMs) on the middleware used in cloud data centres. Additionally, they track measurements and metrics that reflect the operational efficiency of the data centre. Instances of the CACTOS Cloud Infrastructure Models are constructed and maintained by CactoScale. CactoOpt uses the captured models as input for its optimisations. This document gives an overview on the developed models and how they are utilised in the context of a holistic integration process. It relates to other deliverables by integrating the information on CactoScale’s runtime monitoring (D4.2 Preliminary offline trace analysis), CactoOpt’s topology optimisation algorithms (D3.1 Prototype Optimization Model) and the simulative what‐if analyses of CactoSim (D6.1 CactoSim Simulation Framework Initial Prototype) for data centres. Furthermore, recent (D5.2.1 CACTOS Toolkit Version 1) and planned releases of the CACTOS toolkit (D5.2.2 CACTOS Toolkit Version 2) and the licensing models proposed for the individual CACTOS tools are outlined. The feature scope and integration of these features has served as the foundation for the requirements analysis of the developed integration methodology. The current iteration of the CACTOS Cloud Infrastructure Models capture all essential characteristics required to support an integration of current and planned features in all toolkits. Future iterations will improve the usability of the developed models and extend them to address newly identified requirements. In addition, this document describes the development process of the toolkits and the infrastructure used throughout the CACTOS project. The document discusses the setup of CACTOS’ development and build infrastructure and sketches the chosen architecture for the infrastructure. A holistic development process for both CACTOS Runtime Toolkit and the CACTOS Prediction Toolkit was chosen in order to facilitate early as well as Continuous Integration throughout and beyond the project’s life cycle. The build infrastructure was set up following the principle of Continuous Integration and allows for continued development and integration of all tools developed in the CACTOS projects, as well as the tools that they build upon. Finally, the document discusses different licensing models for the release of both toolkits. In line with the effort to keep the results of the CACTOS project open for further development and use by the Open Source community, this document proposes to release all major project contributions under the Eclipse Public License Version 1.

Infrastructure as a Service (IaaS) cloud data centres enable customers to run arbitrary software systems on virtualised infrastructure. In contrast to Software or Platform as a Service approaches, customers do not need to adapt the design of their applications to be cloud-compatible. At the same time, they can benefit from easy scalability and pay-as-you-go models. Customers do not pay for dedicated physical machines. Rather, they are able to request Virtual Machines (VM) with varying characteristics, such as processing speed or memory size. Data centre providers can assign the VMs of multiple customers within their data centre to physical machines. If the VMs are deployed in a manner where the Quality of Service (QoS) of all customers is upheld, the data centre provider benefits from drastically larger economy of scale when compared to traditional one-customer-per-server hosting. The efficient utilisation of the underlying physical infrastructure including management and topology optimisation determines the costs and ultimately the business success for data centre operators. The CACTOS project develops an integrated solution for runtime monitoring, optimisation and prediction. The solution supports data centre providers in data centre management and planning. CACTOS consists of two toolkits: • The CACTOS Runtime Toolkit facilitates automated resource scheduling and optimisation for IaaS data centres. • The CACTOS Prediction Toolkit enables what-if analyses including effects caused by automated resource optimisation based on existing or planned data centre topologies. The CACTOS Runtime Toolkit collects data on a distributed data centre as input to scheduling and optimisation algorithms. Up-to-date load and topology measurements are essential for runtime monitoring, data collection and optimisation. The monitoring and data collection infrastructure introduces unavoidable load in the data centre. The benefit gained by using an automated monitoring and optimisation framework such as the CACTOS Runtime Toolkit strongly depends on the amount of this additional load. The CACTOS Prediction Toolkit requires resources to simulate the behaviour of a data centre. The size and complexity of the simulated data centre influences the feasibility of such a simulative analysis. If the simulative analysis takes a brief amount of time, the data centre planner can quickly account for the results of the simulation and adjust his plans accordingly. This document presents an evaluation methodology for the CACTOS Toolkits as established in (D5.1 Model Integration Method and Supporting Tooling) and (D5.2.1 CACTOS Toolkit Version 1). The evaluation focuses on performance and scalability of the CACTOS Runtime Toolkits. The evaluation approach is driven by the use-case specific requirements for the scientific computing use case of the University of Ulm (c.f. (D7.3.1 Validation Goals and Metrics), (D7.4.1 Validation and Result Analysis)) and Flexiant’s business analytics IaaS hosting use case. For an overview of the use cases, please refer to (D7.1 Scenario Requirements on Context-Aware Topology Optimisation and Simulation) and (D7.4.1 Validation and Result Analysis). The application of the evaluation methodology presented in this document will be outlined in (D5.5 Performance Evaluation of the CACTOS Toolkit on a Small Cloud Testbed). The use case brought into the project by PlayGen will be included in this evaluation. This document closely relates to the documents (D7.3.1 Validation Goals and Metrics) and (D7.4.1 Validation and Result Analysis). These two documents outline goals and results of a practical validation of the CACTOS Runtime Toolkit against the specific goals of each use case. Their focus is on an evaluation in small-scale testbeds and on use-case specific benefit analyses. This document outlines an evaluation methodology that is concerned with the applicability of CACTOS to different testbeds with respect to the performance of the CACTOS tools.

This deliverable describes the final version of the optimization model and algorithms implemented in CactoOpt. The model and algorithms include description of the implemented autoscaling algorithms, their integration with the CACTOS toolkits, and related performance results. In addition, the document describes research results obtained within CACTOS. There are five optimization capabilities of CactoOpt that can be performed on the logical (software) level of data center management: initial placement of virtual machines, migration of virtual machines, shut down of physical machines for energy savings, horizontal scaling, and vertical scaling. Using these four actuators, CactoOpt optimizes the power, performance, and cost tradeoffs of applications running on CACTOS enabled datacenters. The four main actuators enable CactoOpt to optimize for a wide range of scenarios including consolidation, and load balancing. This document elaborates the advances within CactoOpt since D3.3. This includes the improvements in the optimization models, and a thorough description of the new vertical scaling algorithms, horizontal scaling algorithms, fault-tolerant scheduling algorithms, and power capping and management, as well as, the interplay of all optimization capabilities.

This deliverable provides supporting documentation for the official deliverable D6.1, the initial release of the CactoSim simulation framework. It presents the reader with the scope of the deliverable, initial requirements and architectural design for CactoSim. Updated requirements are given, and the foundations that CactoSim are built upon are described. A description of the architecture of the CactoSim V1.0 release is given, and this leads into a description of the graphical user interface by which users can interact with the tool. Provisioning is described, as well as licensing information. Finally, a feature description is given for CactoSim, with an overview of planned future releases also described.

Since the arrival of cloud computing, a significant amount of research has been and continues to be carried out towards the creation of efficient optimisation strategies for meeting certain optimisation goals such as energy efficiency, resource consolidation or performance improvement within virtualised data centres. However, investigating whether specific optimisation algorithms can achieve the desired function in a production environment, and investigating how well they operate are quite complex tasks. Untested optimisation rules typically cannot be directly deployed in the production system, instead requiring manual test-bed experiments. This technique can be prohibitively costly, time consuming and cannot always account for scale and other constraints. This work presents a design-time optimisation evaluation solution based on discrete event simulation for cloud computing. By using a simulation toolkit (CactoSim) coupled with a runtime optimisation toolkit (CactoOpt), a cloud architect is able to create a direct replica model of the data centre production environment and then run simulations which take into account optimisation strategies. Results produced by such simulations can be used to estimate the optimisation algorithm performance under various conditions. In order to test the CactoSim and CactoOpt integration concept, a validation process has been performed on two different scenarios. The first scenario investigates the VM placement algorithm performance within a simulated testbed when admitting new VMs into the system. The second scenario analyses consolidation optimisation strategy impact on resource utilisation, with the objective being to free up nodes towards the goal of energy saving. This deliverable represents the initial part of two iterative pieces of work.

This document outlines a framework for the cloud workload and application models used in CactoOpt, the CACTOS infrastructure optimisation tool, and presents initial prototypes for cloud application behaviour models. The purpose of this deliverable is to demonstrate some of the prediction models built for different cloud workloads, and illustrate how they are integrated with the application and component models used in infrastructure and workload deployment optimization. For prediction modelling we give special focus to cloud application user behaviour modelling, including, e.g., workload burstiness and request arrival pattern modelling. To place this work in context, we also present a framework for application and infrastructure modelling focused on translation of workload and application behaviour to infrastructure load.

Den forskning som bedrivs inom logopedi i Sverige idag spänner över ett brett spektrum av medfödda, utvecklingsrelaterade och förvärvade kommunikationsstörningar på grund av röst-, tal- och språksvårigheter samt sväljstörningar. Precis som i andra länder pågår en snabb utveckling både mot en breddning av det urval av områden som det forskas kring och ett ökat fokus på nationellt samarbete. Även om de framsteg som görs inom respektive fält kan mätas genom de forskningsrapporter som publiceras av respektive forskargrupper, finns det ett behov av att granska vilka forskningsinsatser som bedrivs jämte det demografiska sammanhanget. I detta ståndpunktsdokument ger vi en översikt över logopediämnet och den forskning som bedrivs i Sverige. I en kubmodell illustreras hur kommunikations- och sväljstörningar kan beskrivas. Stratifierade (per åldersintervall) uppskattningar av prevalensen av kommunikations- och sväljstörningar ges baserat på 2016 års befolkningsstruktur. Genom att presentera dessa två komponenter identifieras forskningsområden och riktningar inom logopedi som behöver ges ökad uppmärksamhet i det kommande forskningsarbetet.

In Infrastructure as a Service (IaaS) cloud data centres, customers can run their software on the virtualized infrastructure of a data centre. They benefit from easy scalability and pay-as-you-go payment models and are able to request Virtual Machines (VM) with varying properties, such as processing speed or memory size. Data centre providers benefit from consolidation and economy of scales effects if several VMs are deployed on the same physical resources without Quality of Service (QoS) conflicts, e.g. because VMs often idle and rarely use all available resources. The efficient utilisation of the underlying physical infrastructure including management and topology optimisation determines the costs and ultimately the business success for data centre operators. The CACTOS project develops an integrated solution for runtime monitoring, optimisation and predictions. The solution supports data centre providers in managing and planning data centres. CACTOS consists of two toolkits: • The CACTOS Runtime Toolkit facilitates automated resource planning and optimisation for Infrastructure as a Service (IaaS) data centres. • The CACTOS Prediction Toolkit enables what-if analyses including effects caused by automated resource optimisation based on existing or planned data centre topologies. This document provides an overview on both toolkits and their interactions in the completed first iteration step. The focus of this deliverable is on describing the CACTOS Runtime Toolkit but was extended to give a holistic view and cover the CACTOS Prediction Toolkit as well. The CACTOS Runtime toolkit consists of independent tools for cloud infrastructure analytics and optimisation. This document describes the purpose and features of the tools as well as utilised base technology and provided interfaces. The analytics-oriented tool, CactoScale, provides already an automated extraction of central infrastructure information and monitoring of a running data centre. The optimisation-oriented tool, CactoOpt, can perform optimisation operations on the basis of the extracted information. However, the execution of optimisation operations on cloud middleware requires manual effort. This document describes the provisioning of both toolkits within a data centre and enables testing and running the approach in an own data centre. Exemplary use cases show the applicability and how important tasks are realized in the toolkit. This document delineates how the tools that were developed as part of the individual deliverables for CactoOpt (D3.1 Prototype Optimization Model), for CactoScale (D4.1 Data Collection Framework) and CactoSim (D6.1 CactoSim Simulation Framework Initial Prototype) are integrated into the CACTOS Runtime Toolkit and the CACTOS Prediction Toolkit. Based upon the integration implementation presented in this document, (D5.1 Model Integration Method and Supporting Tooling) will outline the integration methodology that is applied in the toolkits discussed by this deliverable. Exemplary use cases presented in this document were motivated by one of the CACTOS testbeds that is outlined in (D7.2.1 Physical Testbed). The current version of the CACTOS Runtime Toolkit requires some manual interaction with the data centre operator to realise the optimisations. As part of the deployment of the CACTOS Runtime Toolkit in a small-scale testbed (D5.3 Operational Small Scale Cloud Testbed Managed by the CACTOS Toolkit) the integration will be fully automated. In line with the effort of promoting and enabling continued development of the CACTOS toolkits, we released both toolkits under the licensing terms of the Eclipse Public License Version 1. Deliverable (D5.1 Model Integration Method and Supporting Tooling) will contain an in-depth evaluation of different licensing models and the rationale for opting with for the proposed licensing model.

This deliverable outlines a first prototype version of the optimisation model used in CactoOpt, the CACTOS infrastructure optimisation tool. The purpose of this deliverable is to demonstrate interfacing of the optimisation model with preliminary characterization templates describing workloads and infrastructures, i.e. model representations of cloud application workloads (describing virtual machine deployment, configuration, and load) and the data centre (hardware) resources they are executed on. This deliverable additionally shows the integration of CactoOpt in the overall architecture of CACTOS and discusses use cases and optimisation capabilities supported by the first CactoOpt prototype. CactoOpt is designed using a sensor-actuator model where the optimisation engine’s view of the surrounding world is captured in a set of infrastructure topology and load models (sensors) and the actions the optimisation engine can use to affect data centre resources (actuators) are represented using an optimisation plan language (describing a set of infrastructure actions recommended to optimise data centre layout and operation). The model does not assume that all recommended optimisation actions are immediately taken, but rather views these as a set of recommendations the optimizer gives to an external party (e.g., a virtualisation middleware integration implementation or a systems administrator) as part of a greater optimisation plan. This document describes CACTOS deliverable D3.1 – a prototype optimisation model designed for use in CactoOpt. As described in this document, the CactoOpt tool is one of the three main tools in the CACTOS toolkit and this document is related primarily to two other CACTOS year 1 deliverables: CACTOS deliverable D5.2.1 - CACTOS Toolkit Version 1 (2014) that describes the overall design and architecture of the toolkit, and CACTOS deliverable D5.1 - Model Integration and Supporting Tooling (2014) that details the construction of infrastructure topology and workload models and the integration of the different tools in the toolkit based on these models.

This deliverable describes an enhanced version of the optimization model that features predictive capabilities. The purpose of this deliverable is to demonstrate how the enhanced model and advanced optimization algorithms support the optimization of a data center configuration. Predictive optimization capabilities of CactoOpt mainly support three optimization activities that can be performed on the logical (software) level of data center management: initial placement of virtual machines, migration of virtual machines, and vertical scaling. To deliver against these capabilities two software components were implemented: Workload Analysis and Classification Tool (WAC) and Application Behaviour Predictor. WAC is a tool that enables a cloud provider to deploy multiple auto-scaling algorithms suitable for different workload types. The tool assigns a workload to an auto-scaler based on the type of the workload, i.e., some auto-scalers can be better for bursty workloads while other auto-scalers can be better for workloads with strong patterns. The application behavior predictor is a tool that utilizes the knowledge about how the workload and the dynamics of the applications changes over time to predict the future state of the application for optimization purposes, e.g., how long will a task run before terminating on a given hardware configuration.