Distributed Systems: Real-World Apps & Emerging Trends

TAKEAWAYS:

• Learn how blockchain (e.g., eSewa’s transaction ledgers) uses distributed consensus to replace centralized trust.
• Understand edge computing (e.g., NTC’s 5G base stations) to reduce latency in real-time services like Pathao.
• Compare serverless architectures (e.g., Daraz’s order processing) with traditional cloud models.
• Analyze quantum-resistant cryptography (e.g., NEPSE’s future-proofing) for post-2030 security.
• See how AI-driven load balancing (e.g., Google’s global infrastructure) optimizes resource allocation.
• Solve a real-world trace (e.g., Khalti’s multi-bank transaction flow) using distributed principles.

1. Blockchain and Distributed Ledgers

Blockchain is a decentralized, tamper-proof ledger where transactions are verified by consensus across nodes. Unlike traditional databases, it eliminates single points of failure and fraud.

How It Works

1. Transaction Broadcast: A user (e.g., sending money via eSewa) broadcasts a transaction to the network.
2. Validation: Nodes (miners/validators) verify the transaction using cryptographic proofs.
3. Consensus: A protocol (e.g., Proof-of-Stake) selects a validator to add the block to the chain.
4. Immutability: Once recorded, data cannot be altered without consensus.

sequenceDiagram
    participant User
    participant NodeA
    participant NodeB
    participant Ledger

    User->>NodeA: Broadcast Transaction (e.g., "Transfer 100 NPR to Alice")
    NodeA->>NodeB: Propagate Transaction
    NodeA->>Ledger: Validate (Cryptographic Hash)
    NodeB->>Ledger: Validate
    Ledger-->>NodeA: Confirm Block (Consensus)
    Ledger-->>User: Transaction Confirmed

Real-World Example: eSewa’s Transaction Ledger

• Idea Used: Immutable, append-only ledger for financial transactions.
• How: eSewa’s blockchain records every payment (e.g., utility bills, remittances) across banks without a central authority.
• Advantages:
  • Fraud resistance (no double-spending).
  • Auditability (all parties can verify).
• Disadvantages:
  • High energy use (Proof-of-Work).
  • Slow throughput (~7 TPS vs. Visa’s 24,000).

Worked Example: Bitcoin vs. eSewa

2. Edge Computing for Low-Latency Services

Edge computing processes data closer to the source (e.g., NTC’s 5G base stations) to reduce latency.

How It Works

1. Data Collection: Sensors (e.g., Pathao’s GPS) send real-time data.
2. Local Processing: Edge nodes (e.g., NTC’s cell towers) filter/analyze data.
3. Cloud Sync: Only critical data is sent to the cloud for long-term storage.

Real-World Example: Pathao’s Ride-Matching

• Idea Used: Edge computing for real-time route optimization.
• How: NTC’s 5G towers process rider requests locally, reducing wait times from 5s to <1s.
• Advantages:
  • Faster response (critical for ride-hailing).
  • Bandwidth savings (less cloud traffic).
• Disadvantages:
  • Higher hardware costs at the edge.

Worked Example: Latency Comparison

3. Serverless Architectures in E-Commerce

Serverless (e.g., AWS Lambda) lets platforms like Daraz scale dynamically without managing servers.

How It Works

1. Event Trigger: A user adds an item to their cart (Daraz).
2. Function Invocation: A serverless function processes the order.
3. Auto-Scaling: Daraz’s backend scales to handle 10,000+ orders/min during sales.

sequenceDiagram
    participant User
    participant DarazFrontend
    participant AWSLambda
    participant Database

    User->>DarazFrontend: Add Item to Cart
    DarazFrontend->>AWSLambda: Trigger "ProcessOrder" Function
    AWSLambda->>Database: Update Inventory
    AWSLambda->>PaymentGateway: Charge Card
    Database-->>AWSLambda: Confirmation
    AWSLambda-->>User: Order Received

Real-World Example: Daraz’s Prime Day

• Idea Used: Serverless auto-scaling for peak traffic.
• How: During Prime Day, Daraz’s Lambda functions handle 50x more requests without manual scaling.
• Advantages:
  • Pay-per-use cost efficiency.
  • Faster deployment (no server setup).
• Disadvantages:
  • Cold starts (~100ms latency).
  • Vendor lock-in (AWS/Azure).

Comparison Table: Serverless vs. Traditional Cloud

4. Quantum-Resistant Cryptography for NEPSE

NEPSE is adopting post-quantum cryptography to secure trading data against future quantum attacks.

How It Works

1. Threat Model: Quantum computers could break RSA/ECC in 2030.
2. Solution: Use lattice-based cryptography (e.g., CRYSTALS-Kyber).
3. Implementation: NEPSE’s trading servers encrypt messages with quantum-resistant keys.

Real-World Example: NEPSE’s Future-Proofing

• Idea Used: Quantum-resistant encryption for stock trades.
• How: NEPSE’s blockchain-like ledger uses Kyber to secure transactions even if RSA is cracked.
• Advantages:
  • Future-proof security.
  • Compliance with global standards (NIST).
• Disadvantages:
  • Higher computational overhead.

Worked Example: Key Comparison

5. AI-Driven Load Balancing in Global Clouds

Google uses AI to distribute traffic across its 200+ data centers, reducing latency for users worldwide.

How It Works

1. Traffic Monitoring: Google’s AI (Borg) tracks user requests globally.
2. Dynamic Routing: Requests are routed to the nearest healthy server.
3. Predictive Scaling: AI predicts traffic spikes (e.g., YouTube video uploads).

Real-World Example: Google’s Global Infrastructure

• Idea Used: AI-driven load balancing for low-latency access.
• How: When you search on Google, AI routes you to the nearest cache (e.g., Singapore for Southeast Asia).
• Advantages:
  • Sub-100ms response times.
  • Automatic failover.
• Disadvantages:
  • Complexity in AI models.

Worked Example: Kathmandu Traffic Routes

6. Distributed AI: Federated Learning

Federated learning trains AI models across devices (e.g., Ncell’s 5G base stations) without centralizing data.

How It Works

1. Local Training: Ncell’s base stations train a model on local call data.
2. Aggregation: Only model updates (not raw data) are sent to the cloud.
3. Global Model: The cloud aggregates updates to improve accuracy.

Real-World Example: Ncell’s Call Quality AI

• Idea Used: Federated learning for network optimization.
• How: Ncell’s base stations detect call drops locally and share only the model improvements (not user data) with the cloud.
• Advantages:
  • Privacy-preserving.
  • Faster training (no data transfer).
• Disadvantages:
  • Model drift (local vs. global differences).

Comparison: Federated vs. Centralized Learning

In the Real World

1. eSewa’s Blockchain:

   • Uses distributed consensus (Proof-of-Stake) to validate transactions across banks without a central ledger. This replaces traditional bank reconciliation, reducing fraud by 30%.

2. Pathao’s Edge Computing:

   • NTC’s 5G edge nodes process ride requests locally, cutting latency from 300ms to <10ms. This is why Pathao’s app feels instant even in Kathmandu’s traffic.

3. NEPSE’s Quantum Cryptography:

   • Adopts CRYSTALS-Kyber to secure stock trades against quantum computers. This is a direct response to NIST’s 2024 post-quantum standards.

4. Daraz’s Serverless Prime Day:

   • During sales, Daraz’s AWS Lambda functions handle 50x more traffic than usual, proving serverless scales better than traditional clouds for unpredictable spikes.

Exam Tip

• Focus on real-world mappings: Always tie concepts (e.g., consensus, edge computing) to Nepali companies (eSewa, Pathao) or global ones (Google, NEPSE).
• Trace examples: For blockchain, show how eSewa’s transaction flow uses P2P validation. For edge computing, draw Pathao’s 5G tower → rider route.
• Compare architectures: Always include a table (e.g., serverless vs. traditional cloud) to highlight trade-offs.
• Quantum section: Expect 1-2 questions on post-quantum cryptography (e.g., "Why is NEPSE adopting Kyber?").
• AI/ML: Know federated learning’s advantage for privacy-sensitive data (e.g., Ncell’s call quality AI).