- Ruchir Agarwal (Team Leader)
- Akash Kuratkar
- Saumil Shah
- Shravan Vallaban
Introduction to the problem statement
Microservices architecture is a common way to deploy containerized software in the cloud. However, as the number of microservices increases, the management complexity also increases which is due to distributed architecture. Also, Network latency and load balancing are other challenges.
Abstract
Developing a model to determine feasible cloud platform for different types of web-services based on their weightage(Heavy,medium,Light). Training model with parameters like response time, memory utilization, throughput etc. being received from Kubernetes report to predict threshold values for each type services. These predicted threshold values can then be used to determine correct cloud platform for services.
Approach
We plan to develop web services of different weightage like light, medium and heavy and then, we will deploy all these services to the different cloud platforms using docker. Here, we will be using Kubernetes to track performance of all types of services on each platform. After that we will take out reports from kubernetes and will consolidate them into csv and will figure out features(paramters) and labels(prediction values) needed to monitor these services. After tracking all this data for few sometime (few days to weeks) we will use predicted values as threshold value for each type of service considering traffic and latency. Using those threshold values, the model willbe trained and the decision will be made upon which type of service should go to which platform.
Persona
Development team and Dev-ops team.
Dataset Links
We plan to develop our own webservices and tool, to hit the server on each cloud platform and then, use the kubernetes reports for these services as our training data.
Architecture Diagram
