Part of the data-platform path: Big Data Technologies, Cloud Computing for Data Science, Time Series Analysis and Forecasting and Data Engineering and MLOps.
The cloud is not "someone else's computer". It is someone else's computer billed by the second, and that changes what is worth building.
Every technical decision in this course is downstream of one economic fact:
NOTE
Capacity you no longer need can be given back.
That is the whole thing. It is why virtualization matters (you cannot hand back half a physical server), why autoscaling exists, why serverless exists, why storage has six price tiers, and why "just leave it running" is the most expensive habit in the subject.
| The old assumption | What replaces it |
|---|---|
| Buy for the peak, own it for three years | Rent the peak, hourly |
| Capacity is a capital decision | Capacity is an API call |
| An idle server is sunk cost | An idle server is a bill |
| Storage is a disk you bought | Storage is six tiers with penalties |
| Security is a perimeter | Security is identity |
This course names specific products, and the products change. SageMaker Studio replaced SageMaker Notebooks, Stackdriver became Cloud Operations, and the syllabus's product list will read as dated within a few years.
What does not change is underneath: service models, virtualization, storage classes, identity and policy evaluation, the batch/stream split, and the cost arithmetic. Learn those as permanent and the products as examples, and say so in the exam.
There is no cloud account for this repository, and none will be created. Signing up requires a payment card and accepts a billing relationship, which is not something a study repository should do on anyone's behalf.
So this course is the most explicitly split of the ten:
| Runs for real | Documented, NOT EXECUTED |
|---|---|
| IAM's policy evaluation algorithm, implemented and exercised | AWS, Azure, GCP consoles |
| Object-store key semantics — prefixes, no directories, copy-plus-delete | S3, Blob, Cloud Storage |
| All the pricing arithmetic — storage classes, egress, per-TB, per-node-hour | the billing console |
| Hypervisor overcommit, and where it breaks | VMware Workstation |
| A real web server serving a real page over TCP | Apache on a cloud VM |
| A real ETL pipeline into a real columnar warehouse | Glue, Redshift, BigQuery |
| An autoscaling control loop, measured honestly | CloudWatch |
| A real model, and a real AutoML search | SageMaker, Vertex, Azure ML |
| A real HTTP endpoint serving that model, called over the network | a SageMaker endpoint |
That is more than it sounds, because most of what this course teaches is
not proprietary. IAM's three evaluation rules, the fact that an object store
has no directories, and the arithmetic that decides between serverless and
provisioned are all implementable — and all implemented, in
labs/course-13b-cloud/.
Every file that cannot run says *** NOT EXECUTED *** at the top, names
the service it needs, and points at the runnable half.
tools/data-science/run_cloud_labs.py asserts the
marker is still there.
| Unit | Question it answers |
|---|---|
| 1 | What is the cloud, and what are you actually renting? |
| 2 | How is one machine turned into many, and whose machine is it? |
| 3 | Where does the data live, and what does each choice cost? |
| 4 | What does a managed ML platform actually give you? |
| 5 | How do you train, deploy and keep it working? |
Unit 3 is the load-bearing one. Storage decisions are where cloud bills are made and lost, and they are the most examinable arithmetic in the course.
labs/course-13b-cloud/ — the code, and the runner that asserts every figure
these notes quote
data/course-13b-cloud/ — practice datasets, CSV: iam-policies.csv, storage-costs.csv.
Every one was generated from a known truth, so you can score your answer
rather than just produce one; data/README.md lists what each was built
from, data/PRACTICE-QUESTIONS.md sets questions on each with a computed
answer key, and tools/data-science/check_datasets.py proves every one of those
answers against the file.
| From | To | What is shared |
|---|---|---|
| Business Intelligence Tools (BI) | Unit 3, experiment 12 | The same nine-row star schema, imported not copied. ₹10,360 for South is now produced by four engines. |
| Big Data Technologies | Units 3 and 4 | Column projection and partition pruning saved time there; here they save money, on the same mechanism. |
| Machine Learning (ML) | Units 4 and 5 | Identical scikit-learn. The cloud changes the packaging, not the algorithm — and the base-rate argument survives intact. |
| Document Oriented Database (MongoDB) | Unit 3 | Cosmos DB's consistency levels are CAP as a dropdown, with a price per level. |
| Database Management Systems (DBMS) | Unit 3 | A cloud warehouse is columnar, indexless, and billed per byte scanned. Every row of that comparison is a departure from Database Management Systems. |
The syllabus gives a single combined Text / Reference list:
Nayyar, Handbook of Cloud Computing, BPB Publications, 2019 — the closest match to Units 1–3.
Velte, Velte & Elsenpeter, Cloud Computing: A Practical Approach, McGraw Hill.
Gift & Deza, Cloud Computing for Data Analysis, Pragmatic AI Labs — the one that actually addresses Units 4 and 5.
Mishra, Machine Learning in the AWS Cloud: Amazon SageMaker, Wiley.
WATCH OUT
The prescribed list runs 1, 2, 3, 4, 5 with nothing beside the 4 — a title has been lost, and the four books above are items 1, 2, 3 and 5. See review finding D22.
None of the four is free. AWS, Azure and Google Cloud all publish their own documentation and free tiers, and for Units 4 and 5 the vendor documentation is more current than any of these books.
Learn the three service models by what YOU manage, not by examples. The examples change; the boundary does not.
Do the cost arithmetic by hand. 1 TB egress, 1 TB in each storage class, a serverless-vs-provisioned break-even. These are the calculations that get examined.
Learn IAM's three rules and be able to apply them. Explicit deny wins; otherwise allow; otherwise deny. Almost every access question is these three.
Learn one comparison table per unit. IaaS/PaaS/SaaS; the four deployment models; block/file/object; real-time/serverless/batch.
Run the labs. Seven programs, including a real web server, a real ETL into a real warehouse, and a real model served over a real HTTP endpoint.
KEY INSIGHT
Capacity you no longer need can be given back. That is why the cloud exists.
Nothing you forget to switch off will switch itself off. That is why cloud bills surprise people, and it is worth a mark in almost any question about cost.
The NIST definition and the sentence underneath it; the evolution from time-sharing through grid and utility computing; the five essential characteristics, and how to use them as a test; SOA, web services and why REST replaced SOAP; the four-part architecture and the control plane; IaaS, PaaS and SaaS by what the customer manages; continuous delivery, blue/green and canary.
UNIT 2Why virtualization is what makes the cloud possible; partitioning, isolation and encapsulation; type 1 against type 2 hypervisors, and containers; the six types of virtualization; overcommitment measured, and why CPU degrades gracefully while memory falls off a cliff; public, private, community and hybrid; the role of the cloud in data science, and what it does not fix.
UNIT 3Block, file and object storage compared on access unit, sharing and cost; why an object store has no directories and no rename; storage classes, minimum durations and the retrieval fees that reverse the discount; egress and data gravity; backup, archiving, DR and content delivery; key-value databases and their limitations; batch against streaming; cloud data warehouses, bytes scanned, and the break-even.
UNIT 4What the cloud changes about machine learning and what it does not; the benefits and the catch attached to each; AIaaS and GPUaaS as SaaS and IaaS for particular things; managed platforms compared — SageMaker, Azure ML, Vertex AI; the model registry, feature store and experiment tracking; AutoML run for real, what it costs, and the seven things it cannot do.
UNIT 5Choosing a platform — pipeline support, scale-up against scale-out, framework support, pre-tuned services; the six steps and the failure mode at each; the container contract and why /ping must not run the model; real-time, serverless and batch inference with cost figures; monitoring, alarming on the tail, and autoscaling measured honestly; drift, retraining and case studies.
PRACTICEExam-style questions with fully worked solutions.
LABEvery prescribed lab experiment, with code and expected output.