Cloud Computing
From mainframes to microservices

by Adrian Petrescu

Course Overview
1
Day 1 – Public Cloud
Cloud history, AWS platform services, pricing models, live demonstrations, and architectural concepts.
2
Day 2 – Distributed Computing
Scaling strategies, MapReduce, Hadoop ecosystem, and the history of commercial distributed systems.
3
Day 3 – AI/ML Lifecycle
Data processing, feature generation, training methods, inference techniques, and emerging AI applications.
4
Day 4 – ML Ops
Case studies, programming assignments, and demonstrations of VertexAI and Kubeflow pipelines.
Frontrunner for most overused buzzword
“Cloud computing” refers to a shift in the server hardware market towards on-demand, per-hour commodity servers, accessed via automated APIs.
A Brief History of the Cloud
90s
  • In the early computing era, mainframes dominated. Only large enterprises or governments could afford these room-sized machines, and they squeezed every drop of usage from them.
  • A 1999 e-commerce site might invest millions in Sun or Oracle servers, yet still risk catastrophic outages if a single system failed. Like eBay.
Dot-Com Bubble Burst
  • Many 90s startups over-invested in hardware and data centers on speculative growth. When the bubble burst in 2000, those costly assets became “sunk costs” that couldn’t be recouped.
Early 2000s
  • With large-scale investment suddenly much harder to find, more companies starting renting low-cost "VPS" servers from utility-like providers.
  • At the same time, the open-source movement blossomed: free technologies like the LAMP stack lowered the cost of software development.
  • This democratization of infrastructure set the stage for cloud computing
Google: Building the Cloud Before "Cloud" Was Cool
Commodity Hardware Revolution
Google built warehouse-scale computing using affordable PC-class machines instead of expensive servers.
Embracing Failure
Their software expected hardware failures. Data was replicated across servers with automatic failover mechanisms.
At scale, rare events aren't rare.
Pioneering Technologies
MapReduce, Google File System, and BigTable became the building blocks for scalable services.
Internal Cloud
Borg (precursor to Kubernetes) managed workloads efficiently, treating entire warehouses as one big computer.
If Google's first production server resembles a hastily cobbled together amalgam of off-the-shelf computer parts circa 1999, well, that's because it is. Just like Google's original servers at Stanford. If you think this rack is scary, you should see what it replaced. Instead of buying whatever pre-built rack-mount servers Dell, Compaq, and IBM were selling at the time, Google opted to hand-build their server infrastructure themselves. The sagging motherboards and hard drives are literally propped in place on handmade plywood platforms. The power switches are crudely mounted in front, the network cables draped along each side. The poorly routed power connectors snake their way back to generic PC power supplies in the rear.
The Linux Revolution

The Usenet post that started it all

From: torvalds@klaava.Helsinki.FI (Linus Benedict Torvalds) Newsgroups: comp.os.minix Subject: What would you like to see most in minix? Summary: small poll for my new operating system Message-ID: <1991Aug25.205708.9541@klaava.Helsinki.FI> Date: 25 Aug 91 20:57:08 GMT Organization: University of Helsinki Hello everybody out there using minix - I'm doing a (free) operating system (just a hobby, won't be big and professional like gnu) for 386(486) AT clones. This has been brewing since april, and is starting to get ready. I'd like any feedback on things people like/dislike in minix, as my OS resembles it somewhat (same physical layout of the file-system (due to practical reasons) among other things). I've currently ported bash(1.08) and gcc(1.40), and things seem to work. This implies that I'll get something practical within a few months, and I'd like to know what features most people would want. Any suggestions are welcome, but I won't promise I'll implement them :-) Linus (torvalds@kruuna.helsinki.fi) PS. Yes - it's free of any minix code, and it has a multi-threaded fs. It is NOT protable (uses 386 task switching etc), and it probably never will support anything other than AT-harddisks, as that's all I have :-(.

1
1991: Birth of Linux
Linus Torvalds creates a free, Unix-like kernel, out of his college dorm room. The open-source model allows rapid improvement and customization.
2
Early 2000s: Enterprise Adoption
Companies like IBM invest billions in Linux. It becomes the backbone of web servers worldwide.
3
2006-2010: Cloud Foundation
AWS, Azure, and Google Cloud build their infrastructure on Linux. Its flexibility enables on-demand virtual machines.
4
Today: Dominating the Cloud
Linux powers over 90% of public cloud workloads. Its scalability and zero licensing costs make elastic computing economically viable.
Amazon.com in 2006
  • Spiky, yet fairly predictable.
  • Different regions will be offset from each other, but no easy way to pool capacity.
Amazon.com in 2006
  • Spiky, yet fairly predictable.
  • Different regions will be offset from each other, but no easy way to pool capacity.
  • Consistent performance requires overprovisioning by about 60%
Amazon.com in 2006
  • A monthly view tells an even more dramatic story.
  • To stay up during your peaks, you need to overprovision your capacity by ~10x the rest of the time.
  • Even if you can spare the OpEx over this time range, you still need to spend the CapEx.
The Birth of AWS
A perfect storm
The unique scaling needs of Amazon, the hardware innovation from Google, and the rise of Linux virtualization technologies combined to create the perfect market conditions for the first true cloud platform.
Mission
Jeff Bezos approves development of the ideas in Pinkham’s paper and work begins in Cape Town in 2004.
Launch
Launched its first service in March 2006, ended the year with five.
The Golden Age of Cloud Computing
2006: AWS Launch
Amazon launches S3 and EC2, establishing the first true public cloud infrastructure services.
2008: Google enters the fray
Google App Engine launches in 2008, as a platform-level offering that does not meet with immediate success.
2010: Microsoft
Azure begins its life as a defense against the open-source ecosystem that Amazon and Google are enabling.
2011-2014: Service Expansion
Cloud providers rapidly add database, networking, and analytics offerings. Enterprise adoption accelerates.
2015-Present: AI & Serverless
Lambda, machine learning, and specialized hardware transform cloud computing into a comprehensive platform ecosystem.
The Golden Age of Cloud Computing
  • The iconic startups of the early 2000s all built their early product on the cloud.
  • Crucially, the cloud also enabled scaling up without insane capital spikes.
  • Cloud fostered a global market – a startup in India or Brazil had the same access to world-class infrastructure as one in Silicon Valley, leveling the field internationally.
  • By the 2010s, investors and founders had come to expect that infrastructure is not a competitive advantage – it’s a commodity to rent cheaply, like electricity.
Missing the boat
The interesting thing about cloud computing is that we've redefined cloud computing to include everything that we already do... The computer industry is the only industry that is more fashion-driven than women's fashion. Maybe I'm an idiot, but I have no idea what anyone is talking about. What is it? It's complete gibberish. It's insane. When is this idiocy going to stop?
- Larry Ellison, Oracle CEO (2008)

"Maybe I'm an idiot"

A story in three parts.

The Players Today
Amazon Web Services
The pioneer and market leader. AWS offers over 200 services from data centers worldwide.
Microsoft Azure
Leveraging enterprise relationships. Microsoft's cloud platform integrates seamlessly with existing business tools.
Google Cloud Platform
Built on Google's infrastructure. GCP excels in data analytics, AI, and machine learning.
The AWS Service Pyramid
AWS offers a complete spectrum of cloud computing services, each building upon lower levels of abstraction.
1
2
3
1
Software-as-a-Service (SaaS)
Complete applications with minimal management
2
Platform-as-a-Service (PaaS)
Development tools, database management, analytics
3
Infrastructure-as-a-Service (IaaS)
EC2, S3, fundamental building blocks
Core AWS Services: A Snapshot
EC2 (Elastic Compute Cloud)
  • EC2 is the backbone of the entire AWS platform; nearly every other service is either built on top of EC2, or exists to support EC2.
  • You can provision nearly limitless numbers of servers (“instances” in EC2 parlance) in various configurations of CPU cores, RAM, storage, and GPU.
  • These virtual machines run an “Amazon Machine Image” (basically a virtual machine disk image) from a predefined catalogue, or one previously uploaded by the user. This speeds up provisioning times, gives you access to any type of operating system or distribution you want, and helps you centrally manage upgrades/installations/etc.
  • The networking features are layered on top of EC2 through various support services like Amazon VPC, DirectConnect, ELB, etc. As a genera rule, inbound traffic or intra-region traffic is free.
  • Remember: instances are cattle, not pets!
S3 (Simple Storage Service)
  • S3 is an object (or blob) store with virtually unlimited scalability and durability.
  • Amazon’s SLA guarantees 99.999999999% (“11 9s”) durability of objects over a given year:



    “[..] for example, if you store 10,000,000 objects with Amazon S3, you can on average expect to incur a loss of a single object once every 10,000 years.”



    However, this rate so far is purely theoretical. In practice there have been no documented cases of data loss in S3 for anyone.
  • Data ingress from anywhere is free, but data transfer outside of a single AWS region is charged per GB.
  • Includes the concept of multiple “storage tiers” which charge different amounts for different usage patterns.
  • S3 is a popular choice of filesystem for Hadoop-based workflows because of its high local throughput and simple semantics.
RDS (Relational Database Service)
  • Set up, operate, and scale relational databases in the cloud with just a few clicks.
  • At first, it was just a wrapper around the most popular open-source SQL engines of the day (PostgreSQL, MySQL)
  • Over time, it's grown to encompass commercial third-party offerings (Oracle) as well as Amazon's own managed databases (Aurora)
Elastic MapReduce (EMR)
  • EMR is a hosted Hadoop environment that provisions an entire cluster according to cloud-architectural best practices. It removes the notoriously high maintenance burden of a functional Hadoop cluster.
  • In spite of the name, it does not support only MapReduce, but rather the entire Hadoop ecosystem (plus some extras), including Spark and Jupyter.
  • Supports Hadoop-specific autoscaling rules, so your cluster can grow or shrink on its own according to the workload you give it.
  • Fully transparent with respect to its underlying building blocks – you can interact directly with the EC2 instances, VPC rules, S3 buckets, etc. that power it to mold a cluster to any need you have.
EC2 Instances: Cattle, Not Pets
In modern cloud architecture, EC2 instances should be treated as replaceable resources—like cattle—not irreplaceable pets.
  • Pets require individual care and are given unique names.
  • Cattle are numbered, identical, and automatically replaced when needed.
  • This mindset enables true infrastructure automation and resilience.
This philosophy powers auto-scaling groups, immutable infrastructure, and infrastructure-as-code—cornerstone practices in AWS environments.
AWS Pricing Models: Balancing Cost and Flexibility

On-Demand Pricing

A fixed, region-specific price is charged for each instance type, prorated down to the second. Always the same, regardless of utilization or availability. Usually a poor choice for predictable or sustained workloads.

Spot Pricing

Amazon runs a constant auction for unused capacity for each instance type, in each availability zone. While your bid is higher than the current spot price, you get your instance order filled (at the lowest price). As soon as it isn’t anymore, your instance is turned off. Perfect for pre-emptable workloads that can be easily abandoned, resumed, or retried.

Reserved Pricing

Reservations allow you to purchase a discount on the hourly on-demand rate. Reservations are for a specific availability zone and instance type, and cannot be converted (unless it is a special “Convertible RI”) Reserved capacity can be bought and sold on the AWS Marketplace.

Savings Plan

Essentially a more flexible version of Reserved Instances. You’re still purchasing a discount on standard On-Demand rates, but via a term commitment rather than an up-front payment. Less efficient than an RI but more flexible and with better cash flow.

Future Outlook: Trends and Strategic Perspectives in Cloud Computing
The cloud computing industry is evolving into a trillion-dollar market with rapid technological and economic shifts. As we look ahead, several key trends are shaping this landscape.
Multi-Cloud & Hybrid Environments
Enterprises are adopting multi-cloud strategies (averaging 2.6 clouds per company) to avoid vendor lock-in and increase resilience. Better management tools are emerging through platforms like Kubernetes, Anthos, Azure Arc, and HashiCorp. Hybrid solutions like AWS Outposts and edge computing nodes will blur the line between public and private infrastructure.
Industry-Specific Solutions & AI Integration
Cloud providers are moving up the stack with industry-specific clouds for healthcare, finance, and manufacturing. AI will become ubiquitous across all cloud services—from automated data classification to predictive analytics for outages and cost optimization.
New Computing Paradigms
Quantum computing services (AWS Braket, Azure Quantum) will become accessible through cloud platforms. Specialized processing like neuromorphic computing and advanced FPGAs will be delivered as-a-service, continuing the trend of cloud-based access to exotic hardware.
Regulatory & Geopolitical Factors
Data sovereignty laws are forcing providers to build more local regions. Antitrust scrutiny is increasing, with potential remedies like standardized APIs or caps on data egress costs.
Made with