DBATU · Third Year B.Tech Artificial Intelligence & Allied · NEP 2020
Cloud and Distributed AI Systems
26AFAIPC503 · Semester V · 3 credits · Continuous assessment 20 + mid-semester 20 + end-semester 60
Unit 1: Introduction to Cloud and Virtualisation
5 hoursThe vision and definition of cloud computing, its reference model, characteristics, benefits, challenges and history; virtualised environments and the taxonomy of virtualisation techniques, its pros and cons; VMware and Microsoft Hyper-V.
Examined in the 60-mark end-semester paper, and in the mid-semester exam if it is taught by then
- Cloud characteristics
- Cloud reference model
- Types of virtualisation
- Hypervisors
Unit 2: Cloud Computing Architecture
8 hoursIaaS, PaaS and SaaS; public, private, hybrid and community clouds; the economics of the cloud, interoperability, standards, scalability and fault tolerance; storage, database, process, information, integration and testing as a service; capacity planning and scaling cloud infrastructure.
Examined in the 60-mark end-semester paper, and in the mid-semester exam if it is taught by then
- Cloud service models
- Cloud deployment models
- Everything as a service for the enterprise
- Capacity planning and scaling cloud infrastructure
Unit 3: Aneka Cloud Application Platform
6 hoursThe Aneka framework and container; platform abstraction layer, fabric, foundation and application services; building Aneka clouds and their infrastructure and logical organisation; private, public and hybrid deployment modes; cloud programming and management with the Aneka SDK.
Examined in the 60-mark end-semester paper, and in the mid-semester exam if it is taught by then
- Aneka framework and container services
- Aneka deployment modes
- Cloud programming and management with Aneka
Unit 4: Distributed AI Systems
5 hoursFrom classical machine learning to deep learning and LLMs, and the need for distributed AI; compute, storage, networking and workflows in distributed AI; CPU versus GPU, GPU basics and CUDA, multi-GPU communication; data, model and pipeline parallelism; scalability, communication overhead and fault tolerance.
Examined in the 60-mark end-semester paper, and in the mid-semester exam if it is taught by then
- Distributed AI system architecture
- GPUs for AI training and inference
- Data, model and pipeline parallelism
- Challenges in distributed AI systems
Unit 5: Distributed Training of AI Models
5 hoursWhy distributed training; PyTorch distributed concepts and multi-GPU environments; distributed data parallel and gradient synchronisation; dataset partitioning, distributed sampling and batching; GPU memory limits and mixed-precision training; fully sharded data parallel and how it compares with DDP.
Examined in the 60-mark end-semester paper, and in the mid-semester exam if it is taught by then
- Distributed data parallel training
- Partitioning and sampling training data
- Mixed-precision training and GPU memory
- Fully sharded data parallel
Unit contents are summarised from the official DBATU syllabus . AiCanCode.org is not affiliated with Dr. Babasaheb Ambedkar Technological University (DBATU).