Getting Started

The first task carried out by the Optimisation tool plug-in is to provide the user with a simple and intuitive graphical interface to interact with the optimization engine. In particular, it allows to: i) upload the input files defining the problem instance to optimize, ii) launch, stop and cancel the optimization process, iii) download the results of the optimization process.

The main screenshots of D-SPACE4Cloud Eclipse plug-in are proposed in Figure 1–Figure 10. The plug-in implements a five step wizard whose windows change according to the selected technology and target deployment. In this section we will provide an overview of the plug-in usage when the Spark technology and public cloud deployment are selected. The complete description is reported in the Usage page.

In particular, Figure 1 shows the initial window of the D-SPACE4Cloud. The project window depicts, as an example, a Spark application DTSM model. The optimization tool starts pressing the entry "Optimization Wizard" in the menu.

Figure 1: D-SPACE4Cloud Main Window

In the first step of the wizard (see Figure 2) the user has to specify the number of classes, the DIA target technology and the deployment (public or private). If the public cloud deployment is selected (see Figure 3) the user has to specify also if there is an existing long term contract (e.g., Amazon reserved instances) or not and, in the former case, specify the number of reserved instances. The premium release of D-SPACE4Cloud supports also the use of spot instances (and in this case the fraction, i.e. a number between 0 and 1, of the cluster capacity to be devoted as spot needs to be specified).

Figure 2: Wizard step 1. Selecting DIA technology and target deployment

Figure 3: Public cloud deployment with existing long term contract

Next (see Figure 4), the user has to specify the optimization alternatives selecting, possibly, multiple VM types at different providers candidate for the final deployment. For each VM type, the user needs to select the corresponding DTSM model (see Figure 5) profiled with the service demands expected when the DIA runs on the candidate VM type. The last input of this step, is the DDSM model (see Figure 6), which includes the deployment model of the DIA which will be updated by D-SPACE4Cloud with the optimal solution found. Such model can be processed by the DICER tool to obtain the TOSCA description of the optimal configuration, whose automatic deployment can be obtained through the DICE delivery service.

Figure 4: VM type selection and DICE model specification

Figure 5: DTSM selection

Figure 6: DDSM selection

Note that, in the D-SPACE4Cloud premium version, a machine learning model, which is able to predict the DIA model performance according to its profile characteristics, can be also provided. This significantly speed-up the optimization process providing a very good initial solution for the local search procedure.

The next wizard window allows to input the optimization constraints and it is technology specific. In particular, for Spark and Hadoop MapReduce (see Figure 7), the user can specify the minimum and maximum number of concurrent users, the DIA end users’ think time and the DIA deadline (job penalty cost can be specified only on the private cloud case). Note that, for Spark the minimum and maximum number of users needs to be equal to 1 if JMT and GreatSPN simulation tools are used while dagSim supports any number.

Figure 7: Spark and Hadoop MapReduce optimization constraints

When also the optimization constraints are specified, the end user can press the Finish button. Then the window in Figure 8 is shown and the optimization process starts. When the final solution is obtained, the window in Figure 9 is displayed and the results can be downloaded by selecting the entry "Show Public Results" or "Show Private Results" (according to public or private deployments) from the menu in Figure 1. Note that, the DDSM model selected in the previous steps will be automatically updated while additional files or information can de obtained through the window in Figure 10 like D-SPACE4Cloud start time or the cost of the final solution found. The results window allows also to cancel or restart an optimization process if it failed for any issues and to download the input files and low level output files used by the backend (file format is discussed in DICE Deliverable D3.8).

Figure 8: D-SPACE4Cloud wizard Finish window

Figure 9: Download Window

Figure 10: Results Window