Cloud Computing (IN2107)

Prof. Dr. Michael Gerndt, Mohak Chadha Jianfeng Gu, Srishti Dasgupta, Prof. Amitava Gupta

Organization

• We have a total of 12 available places for our cloud computing seminar.
• Each student will select a seminar topic and submit a paper on that topic.
• We will use hotcrp as the tool for submission of seminar papers.
• To make students more familiar with the scientific peer-reviewing enviornment, each submitted paper will be assigned two fellow students that will review the paper and give feedback.
• Each student will also prepare a 30 minute presentation on their topic.
• All seminar presentations will be held in-person at Frauenchiemsee. Transportation and stay will be covered by the Chair. Food will not be covered which will cost around 60 euros. Students must have a deutschland ticket.
• Presentation tentative dates: Jan 2024 End.

Motivation

Cloud computing is dominating the IT sector due to its on-demand provisioning of resources. It enables highly elastic and scalable application work loads with huge differences in computational and storage demand. The Internet of Things is an excellent use case for clouds as it has a varying amount of incoming sensor data creating huge storage requirements. Accesses to the data might be based on short running functions but also compute intensive analytics jobs are run on the data. Other applications follow the cloud native model where the application is decomposed into microservices which are deployed based on backend services of the cloud provider into VMs or Kubernetes clusters. In combination with the microservices more and more applications also use function-as-a-service offerings of the cloud providers. Running these complex parallel and distributed applications efficiently on the large scale cloud infrastructure, even in combination with distributed resource on the edge, poses a lot of challenges on the management of the applications as well as of the infrastructure. 

The topics that will be covered in this instance of the seminar are: 

• Data Driven Simulation of Physical Systems (Supervisor: Prof. Amitava Gupta)
• Visualization Platforms for Digital Twins (Supervisor: Prof. Amitava Gupta)
• Survey on Scheduling in Hybrid Cloud (Supervisor: Srishti Dasgupta) (Paper)
• Towards Standard Kubernetes Scheduling Interfaces for Converged Computing (Supervisor: Srishti Dasgupta) (Paper)
• Constrained Energy-Cost-Aware Workflow Scheduling for Cloud Environment (Supervisor: Srishti Dasgupta) (Paper)
• Optimal Cloud Resource Selection Method Considering Hard and Soft Constraints and Multiple Conflicting Objectives (Supervisor: Srishti Dasgupta) (Paper)
• Deploy Our Own ChatGPT: Enabling Efficient Inference Serving for LLM (Large Language Model) in the Cloud. (Supervisor: Jianfeng Gu) (Info)
• IceCrusher: Alleviating Cold-Start for Serverless Function (Supervisor: Jianfeng Gu) (Info)
• Awesome FaaS-Inference: Optimizing Serverless Computing for Deep Learning Inference. (Supervisor: Jianfeng Gu) (Info)
• FaaS-Issues: Existing Problems in Serverless Computing. (Supervisor: Jianfeng Gu) (Info)
• DL-Profiling: How to Profile Resource Usage of Deep Learning Models. (Supervisor: Jianfeng Gu) (Info)
• Checkpointing in CUDA. (Supervisor: Jianfeng Gu) (Info)
• Understanding Apache TVM. (Supervisor: Jianfeng Gu) (Info)
• Sustainable cloud computing. (Supervisor: Mohak Chadha) (Paper)
• Exploring CXL-based memory in Cloud platforms.  (Supervisor: Mohak Chadha) (Paper)
• Cheaper Serverless Computing with Harvested VMs. (Supervisor: Mohak Chadha) (Paper.)

The students will select material for the presentation, provide a report, and give a presentation of 30 minutes. The management of the seminar is done through Moodle.