Cloud Platforms & Case Studies: AWS, Azure, GCP, Hybrid Clouds

Key Cloud Platforms and Their Architectures

1. Amazon Web Services (AWS)

AWS is the market leader in cloud computing, offering over 200 services including compute (EC2), storage (S3), databases (RDS), and AI (SageMaker). Its architecture follows a multi-region, multi-AZ (Availability Zone) design to ensure high availability and fault tolerance.

Core AWS Services

AWS Global Infrastructure

AWS operates 33 geographic regions, each with multiple Availability Zones (AZs). AZs are isolated locations within a region connected via low-latency links. For example:

• Ncell’s AWS Migration: Ncell (Nepal’s largest telecom) migrated its customer support chatbot and billing system to AWS to handle 10M+ daily transactions. They used:
  • EC2 for scalable backend servers.
  • S3 for storing customer data backups.
  • Lambda to trigger alerts for fraudulent transactions.

2. Microsoft Azure

Azure is Microsoft’s cloud platform, tightly integrated with Windows Server, Active Directory, and Office 365. It is preferred by enterprises using Microsoft products (e.g., banks, government agencies).

Key Azure Services

Hybrid Cloud Example: Daraz (Alibaba Group)

Daraz, Nepal’s largest e-commerce platform, uses a hybrid cloud model:

• Public Cloud (Azure): Hosts scalable web apps (React.js frontend, Node.js backend) and AI recommendation engines.
• Private Cloud (On-Prem): Stores sensitive customer data (payment records) in a high-security data center in Kathmandu.
• Connectivity: Uses Azure ExpressRoute (private fiber link) to sync inventory between cloud and on-prem systems.

3. Google Cloud Platform (GCP)

GCP is known for its AI/ML tools (TensorFlow, Vertex AI), global network (low-latency backbone), and Kubernetes expertise. It powers YouTube, Google Search, and Android backend services.

GCP’s Unique Features

Case Study: NEPSE’s Cloud Migration

Nepal Stock Exchange (NEPSE) migrated its trading platform to GCP to:

1. Reduce latency: Used Google Cloud’s Singapore region (closest to Nepal) to process trades in <30ms.
2. Scale for Diwali peak: Auto-scaled Compute Engine instances during high-volume trading days.
3. Cost savings: Paid 30% less than on-prem servers by using preemptible VMs for batch jobs.

sequenceDiagram
    participant User as Trader (Nepal)
    participant NEPSE as NEPSE Website
    participant GCP as Google Cloud (Singapore)
    participant DB as Firestore (Multi-Region)

    User->>NEPSE: Places order via mobile app
    NEPSE->>GCP: Routes request to Singapore region
    GCP->>DB: Validates order in Firestore
    DB-->>GCP: Returns confirmation
    GCP->>NEPSE: Updates trade status
    NEPSE-->>User: Shows confirmation
    Note over GCP,DB: All data replicated across 3 regions for HA.

Cloud Deployment Models Compared

Not all clouds are equal. Choose based on security, cost, and scalability needs.

Cost Optimization Strategies

Cloud costs can spiral if unchecked. Here’s how to save money:

1. Right-Sizing Resources

• Example: A Daraz order-processing server was over-provisioned with 8 vCPUs but only used 20% CPU. Resizing to 2 vCPUs saved $1,200/month.
• Tools: AWS Trusted Advisor, Azure Cost Management, GCP’s Recommendations.

2. Spot/Preemptible Instances

• Spot Instances (AWS): Bid for unused capacity (up to 90% discount).
  • Use case: Batch processing (e.g., NTC’s nightly traffic data analysis).
• Preemptible VMs (GCP): Cheaper VMs that can be terminated with 30s notice.

3. Reserved Instances

• 1- or 3-year commitments for steady workloads (e.g., Ncell’s CRM servers).
• Savings: Up to 75% vs. on-demand pricing.

4. Storage Tiering

Worked Example: NTC’s Traffic Data Storage

• Problem: NTC stores 10TB of GPS traffic data but only queries 1TB/month.
• Solution: Moved 9TB to Glacier Deep Archive, saving $2,400/year.

Security and Compliance in the Cloud

Cloud providers share responsibility with customers. AWS/Azure/GCP handle physical security, networking, and hypervisor security, but you own:

• Data encryption (KMS, customer-managed keys).
• IAM policies (least-privilege access).
• Application security (OWASP Top 10).

Real-World Compliance Challenges

In the Real World

1. eSewa (Nepal):

   • Idea Used: Serverless architecture (AWS Lambda).
   • How: eSewa’s payment gateway processes 50,000 transactions/hour during festivals. Instead of running 24/7 servers, it uses Lambda functions triggered by API calls. This reduces costs by 60% compared to always-on VMs.

2. Pathao (Ride-Hailing):

   • Idea Used: Multi-region deployment (AWS + GCP).
   • How: Pathao’s backend runs in AWS (Singapore) for Southeast Asia and GCP (Iowa) for US operations. During the 2023 monsoon season, AWS’s Mumbai region faced outages, but traffic rerouted to GCP’s Tokyo region with <100ms latency.

3. NTC’s Smart Traffic System:

   • Idea Used: Edge Computing (AWS IoT Greengrass).
   • How: NTC’s 500+ traffic cameras in Kathmandu don’t send raw video to the cloud. Instead, Greengrass processes frames locally, sending only anomaly alerts (e.g., accidents) to AWS. This reduces bandwidth by 95% and cuts costs by $8,000/month.

Exam Tip

This unit is heavily tested on:

1. Platform Comparisons: Be ready to compare AWS, Azure, and GCP on pricing, services, and use cases (e.g., "Why would a Nepali bank choose Azure over AWS?").
2. Case Study Analysis: Expect questions like:
   • "How does Daraz’s hybrid cloud reduce latency for Kathmandu users?"
   • "What security controls would you implement for NMB Bank’s cloud migration?"
3. Cost Calculations: Practice estimating costs for scenarios like:
   • "A startup runs 5 t2.micro EC2 instances 24/7 for a month. What’s the AWS bill?" (Answer: $36.50).
4. Diagrams: Always draw architecture diagrams (e.g., hybrid cloud, VPC) in exams. Use mermaid syntax if allowed.
5. Compliance Shortcuts: Memorize these mappings:
   • PCI-DSS → Payment data (e.g., eSewa).
   • GDPR → EU customer data (e.g., Ncell’s international calls).
   • SOX → Financial audits (e.g., NMB Bank).

Pro Tip: For 10 marks, describe a real-world case study (e.g., Ncell’s AWS migration) with:

• Problem → Solution → Tech used → Outcome.
• Example:

  "NTC’s traffic cameras generated 10TB/month of video data. They used AWS IoT Greengrass to process data at the edge, reducing cloud costs by $8,000/month while maintaining real-time alerts."