Proactive and Polymorphic Adaptation of Multi-Cloud Deployments

Yeah, we have the last moment chance. So I'm not alone together with me. Nevi team leader in, in our company.

And yeah, we, we want to, to make our session more attractive and present, also live demo of the, of the platform of the multi-cloud platform, so that the reason that we are together. Yeah. And I will start with the few slides and then ALA will present the, and the live demo and also make introduction. So let me share my screen probably as we already tested, video will disappear, but we have agreed. The most important thing is, is on the slide. Yeah.

So today we want to tell you a bit about the project, which in which we participate, it is the melodic plot project melodic platform, which transform into the, the morphic platform morph morphic project, which It looks like, like we don't receive the videos as video. Here we go. Perfect. Okay. So we see the slides we can continue from here. Thank you. Yeah. Yeah. Unfortunately, yeah.

As I said, the slide is more important than the video I'm using teams on the browser cause on Linux platform issue with using desktop applications. So that's the reason, but as I said, the most important thing on the, on the slides. Yeah. So shall I continue? Yeah.

Okay, great. Okay. So today we want very briefly tell about the, the melodic platform, which is fully open source, single universal platform for optimized deployment and management application in the cloud. It was created in the horizon 20, 20 melodic project. And now we for download and to use both for the commercial proposals, for the individual proposals, as well as for the, for the research, we are continuing developing the, the platform. It is also deployed to the, some of the commercial customers.

And the main goal of the platform is to optimize and automate the deployment of the application in the multi-cloud environment. Yeah.

As I said, the application is fully open source, so you can download and use. I will tell a bit about then at the end. So first thing why to use that platform like melodic and first, and probably the most important reason is that it is the simplest and easiest way to use the multi-cloud approach. At least known for me, it is also the unified way to deploy virtual machines, containers, serverless, and big data framework to the different cloud providers. I will show during the live presentation. Yeah.

How easy it is to deploy one application to the different cloud providers and deployment is fully automatic and even more, the cloud resources are automatically optimized using optimization. The first is to model the application. And for that, we have developed the Kamel cloud application modeling and execution language, which is somehow similar to Tosca. So it is some kind of the metal language over the infrastructure as a code language, it tells for the modeling components, connections, security, and also for the modeling requirements for the infrastructure.

So we do not model usually the direct infrastructure elements, but we are modeling the requirements minimum number of course, maximum number of course, minimum number of memory and so on. And based on that, we can construct the requirements constraint and so-called utility function, which allows us to optimize the resources. I will tell a bit more about that on the next slide, but thanks to that, it is possible to make a unified way of describing both application and infrastructure in the cloud diagnostic way. Probably the most unique thing in melodic is this best deployment.

So how melodic knows what is the best deployment before that formation melodic is collecting metric from the running application. The dedicated metric collection system is built into the, into the melodic and able to operate in the multi-cloud environment. So melodic is collecting different type of the metrics from the different cloud providers. And these metrics are used to calculate the utility function of the particular application.

This utility function is concentrate on the business value of the application, for example, the average response time to the user or average time or estimated time to finish the given job and so on. So thanks to that. We are able to, we are able to, oops, can you hear me? And Yeah, we, we can hear, but we've lost the picture again. We've love. Okay. Now it's better. It's back now.

Okay, great. Great. So I will continue just a few slides and we will switch to the life demonstration. Yeah. So very briefly about the best deployment. So we can optimize the, the different factors, which impact of the deployment. So the cost performance of the ability and to find the best possible deployment configuration. So we can consider melodic as your smart, automatic DevOps who will choose the best configuration from your point of view.

Yes, we are running out time. I will go briefly through for this slide, maybe. So how melodic works. Yeah. The first two steps are to model and define the application.

As I said, in the camera language and then melodic calculate the initial deployment based on the initial values of the parameters. And based on that configuration, the application is automatically deployed to the different cloud providers. Then melodic started to collecting metrics and mentoring application. If the metrics exceeds threshold, then melodic is started to looking for the new, new, optimal solution after founding the new optimal solution, the application is reconfigured automatically. We will show that live so you can see how it looks and how it works in, in practice. Yeah.

Before that, I will also go very quickly for the one use case. ALA will present another one. So you can see how, how it looks in practice. This use case is based on the AI investments. It is the Polish company which create the complete platform for the investment portfolio optimization using machine learning models. And their typical business call is to train for example, 50 prediction models in one hour using as minimal numbers of resources as possible and how it looks in practice.

Yeah, the investments Analyst Analyst Analyst in AI investments starts to train models at first on permission, infrastructure is used cause it's already bought already. And the cheapest one, at least in terms of the marginal cost, but usually the on premises resources are not enough. So melodic is calculating the estimated time to finish training and add necessary resources, selecting the most efficient cloud providers. If it's still too long, then additional resources are added.

If the train, if there are enough resources added, then the results of the training are store and all of the cloud resources are removed automatically. Yeah. So that's very simple use case, but it is very beneficial at least for the AI investments, based on the AI investments calculation, it saves around the 175,000 us dollars in three years perspective, thanks to the optimization of the, of the cloud resources. And now I think due to the time limitation, we should switch to the live demo and allow will present another application using the big data framework spark. Okay. Okay.

So now I share my screen and during my part of presentation, I would like to show how melodic works in practice. I will perform deployment of spark based application. As pav. I already mentioned, we will monitor application metrics and observe frequent configuration process, which is done by melodic for reasons of optimization. My platform is already in start on virtual machine and it is up and running. I'm locked in. I have already configured my platform. This process. I will describe a bit later. Now we can go to the deployment bookmark today.

I would like to deploy genome application before deployment. We need to model our application with its requirements in camera model, which is human understandable and editable form. After that such model is transformed by dedicated tool to Xi format form understandable formal logic. I have already attached such X semi file here. So now it is stored in database and I can go to the next step when I need to choose which application I want to deploy and which cloud providers I want to use.

Now, I would like to focus on deployment on AWS. So I choose this cloud definition. Thanks to that. Melodic has credentials to this provider because definition contains credentials provided by me a bit later, a bit before this step. And after that, I can go to the last step where starting deployment is available after starting the process. In a minute, we are moved to the deployment process view where we can observe the progress of that. In the meantime, I would like to briefly describe which is being deployed by melodic now genome.

So the application deployed by melodic during this demo is a big data application, which performs some calculations and saves results in I AWS as free bucket. So we need to provide developers, credentials, toss, and it was down by me during uploading of semi file. With model of application genome's performance is managed by spark in general application. We use spark as platform for big data operations, which are performed parallel on many machines and managed by one machine named spark master spark master is labeled by default on melodic platform.

Melodic creates proper number of spark workers as vegetable machines considered our requirement from camera model. Thanks to measurements of application metrics. Melodic makes a decision about creating additional instances with workers or about leading ones, spark device, all calculations names, tasks between workers in order to optimize application performance and cost.

Now, I would like to briefly describe the configuration process of tic platform, which was down by me also just before this demo because of optimization of time of our presentation. So in cloud definitions part in provider settings, we have settings for all providers, which we would like to use, and we need to provide such values as for example, S and properties for providers. And it is really required because they are used in contact with providers.

For example, by creating virtual machines instances, online environment, I have already defined two cloud definitions, but also it is available to create a site definition, for example, for GCP or Azure. Okay, so now please let me come back to our deployment process. Fe offers is the first step of deployment process. We have information about current total number of offers from previously selected providers. So in our case from AWS, from these offers, melanic will choose the best solution for worker component. After choosing this box, we are directed to view of all currently available offers.

There are clouds with my definition of cloud fors. So with properties and credentials inside, then we can series of hardware with information about course Ram discs. And after that list of available locations where our visual machines could be located and images list here, we can see only private images, but of course all public images are available for us. And now I back again to our process and we can see that the next step of process in generating constraint problem constraint problem is generated based on our requirements defined in camera model and simple process view.

There are visualized all variables from constraint problem with the dominant values. So in case of general application, we have worker cardinality worker course and provider for worker data are shown after on this box. And here are presented these of variables with additional information about component, eh, type domain and eh, type domain. After that, we can see utility formula. It is used for measure utility of each possible solutions and choose the best one list of constants with types and values. They are created from user requirements and are used in calculations. Here.

We can see, for example, minimum and maximum value for cardinality of spark worker. So in this deployment, we would like to have from one, two maximum 10 workers, and we can see the same type of restriction for a number, of course. And we use such such constraint in order to optimize the cost of, of the, of the whole deployment process and last limit here, list of metrics with the types and values. They describe current performance of this application. Thanks to them.

Melodic can make a decision about regarding the reconfiguration process, which means creating new, additional instances or not fully used ones. Thanks to metrics. Melodic can do the most important task, which is cross optimization. And now we are back to process view when Costway problem is generated. It is time for rezoning, melodic finds here the best and most profitable solution for the problem defined by us. When rezoning is completed, we can observe information about calculated solution, so formula and values for each variables.

In that case, we have one worker with four course and provider for spark worker from zero zero index, which means AWS. And the next step in process deployment is deploying Helo, performs operations based on calculated solution. So this solution is deployed for each application component, melodic creates proper instances, remains them or do it. And now we can see that still our solution is being deployed. And in the meantime, also we can observe the advanced view of our process by using this button. And now we are directed to Kaunda. Kaunda is a tool for modeling processes in BPM and standard.

And for management of them, I need to log in by the same credentials as for the home logic platform. And I need to choose the process, which I would like to monitor. And here we have only one process our deployment process, and we can observe each step of, of them. It is more, this is you for more technical users, and it could be useful for example, during diagnostic of some problem, because we can see here all variables with values from, from our process. And we can see here that now we are in this step, which means waiting for notification, and this is close to the end of the process.

Also, I can refresh dedicated view of our process. Okay? So we are still waiting for notification. And now we can, we can back to our process view when our virtual machines are created, we will see the result in your application, mark. And I suppose that we, our, we can already see the virtual machine. Yes. And now we logic performs some installation steps on the virtual machine, but the virtual machine is ready to use. And it was already created by melodic in this deployment process. And we can see information about this virtual one machine.

So the information about IP addresses about provider and the location of, of it, what is more we have here button for web SSH connection, which is really useful in testing process. Also after successful deployment of spark application, we can go to Grafana. Grafana is tool for monitoring, displaying statistics and metrics, and we can use it for monitoring performance of applications deployed by melodic.

Again, I need log. Each application has own metrics and own parameters to control. So we need to create the dedicated graph dashboard for each of them. Also genome application has on graph settings and we can see these settings here and now our application is still is being started. So we can see now the exact data of our, our deployment, but in a minute, we should see the real values. So here we can observe number of instances. And now we can, we have one, one instance, one worker instance and spark master is built into the modic platform. So we observe only, only workers.

And also we will see the, the time left because in our common model, we, we have indicated indicated time and it was, and it is equal to one hour and we cannot observe here how many minutes left from this time period also based on current perform melodic will calculate it, estimated time left for our application and Visible if it would, when you find it's somehow interested. Then as I said, you can download melodic from, from this link it's released under Moula public license. So it can be freely used.

Also, you can join our open source community to develop the melodic. And also you can follow us on the social media, on LinkedIn, Twitter, Facebook, and to our website. We are very extensively work on the multi-cloud deployments and optimization of the cloud resources. So I hope you will find something interesting. Thank you very much. Thank You, power. And thank you both of you for, for that presentation. I appreciate you being with us today and I hope everyone here enjoyed it and also online, but for now, thank you very much.