Data Center Architecture: Design, Components & Challenges

TAKEAWAYS:

• A data center is a centralized facility housing servers, storage, networking, and power systems to deliver scalable, secure, and reliable IT services.
• Physical components include servers, racks, UPS, cooling systems, and fire suppression, while logical layers (network, storage, compute) follow a tiered architecture (Tier 1–4).
• Design considerations prioritize scalability (modular expansion), redundancy (N+1 power/cooling), security (biometrics, firewalls), and efficiency (PUE, cooling strategies).
• Challenges in Nepal’s context include high electricity costs, limited fiber backbone, and monsoon-related power outages—mitigated via hybrid cooling, diesel generators, and cloud bursting.
• Real-world examples: Ncell’s data center uses Tier 3 redundancy for 99.98% uptime; eSewa’s hot-aisle/cold-aisle containment cuts cooling costs by 30%; Daraz’s micro-data centers in Pokhara reduce latency for local users.
• Exam focus: Memorize the 4-tier classification, PUE formula, and disaster recovery strategies (backup sites, snapshots, replication).

1. What Is a Data Center?

A data center (DC) is a physical or virtualized facility that centralizes an organization’s IT operations, including:

• Compute: Servers (physical/virtual) for processing.
• Storage: NAS, SAN, or cloud storage for data.
• Networking: Routers, switches, and firewalls for connectivity.
• Power/Cooling: UPS, generators, and HVAC for reliability.
• Security: Biometrics, CCTV, and access controls.

Why Are Data Centers Critical?

mindmap
  root((Why Data Centers?))
    "Centralized Management"
      "Single point for updates/patches"
      "Unified security policies"
    "Scalability"
      "Add servers/storage on demand"
      "Cloud integration (AWS, Azure)"
    "Disaster Recovery"
      "Backup sites (hot/cold)"
      "Redundant power/cooling"
    "Cost Efficiency"
      "Economies of scale (shared cooling)"
      "Energy monitoring (PUE)"
    "Modern Business Needs"
      "E-commerce (Daraz, Amazon)"
      "Banking (NMB, SBI)"
      "Telecom (Ncell, NTC)"

2. Components of a Data Center

Data centers are built from interdependent layers. Below is a breakdown of physical and logical components:

A. Physical Infrastructure

B. Logical Layers (Tiered Architecture)

Data centers follow a layered model for scalability and fault tolerance. The Uptime Institute classifies them into 4 tiers:

Worked Example: Ncell’s Data Center Tier Ncell’s Kathmandu data center is Tier 3:

• Power: Dual UPS + diesel generator (N+1).
• Cooling: Hot-aisle/cold-aisle containment with CRAC units.
• Redundancy: Dual network paths to ISPs (NTC, WorldLink).
• Uptime: 99.98% (allows 1.6 hours of downtime/year).

3. Key Design Considerations

When designing a data center, engineers focus on 5 critical factors:

A. Scalability

• Modular design: Add racks/servers without downtime.
• Virtualization: Use VMware/Hyper-V to scale compute dynamically.
• Cloud integration: Hybrid model (on-premise + AWS/Azure).

Example: Daraz’s Pokhara micro-data center uses Dell EMC VxRail for scalable VMs to handle local orders.

B. Redundancy

• Power: N+1 (1 extra UPS/generator beyond capacity).
• Cooling: Dual CRAC units with automatic failover.
• Network: Dual ISPs (e.g., NTC + WorldLink).

Worked Example: eSewa’s Redundancy eSewa’s DC has:

• 2 UPS units (each 100kVA) + 1 diesel generator.
• Dual cooling paths: If one CRAC fails, the other takes over.
• Result: Zero downtime during Kathmandu’s 2022 load-shedding.

C. Security

D. Energy Efficiency

Measured by PUE (Power Usage Effectiveness): $$\text{PUE}=\frac{\text{Total Facility Power}}{\text{IT Equipment Power}}$$

• Ideal PUE: 1.2–1.5 (Google’s DC: 1.1).
• Nepal’s Challenge: High PUE (~2.0) due to old cooling systems and power fluctuations.

Example: Ncell’s DC improved PUE from 2.1 → 1.6 by:

1. Switching to liquid cooling.
2. Using free cooling (outside air during winters).
3. Virtualizing 60% of servers.

E. Cooling Strategies

4. Challenges in Data Center Operations (Nepal Context)

Worked Example: NTC’s Fiber Challenge NTC’s Kathmandu–Pokhara fiber faces:

• 30% signal loss during monsoons (due to flooded trenches).
• Solution: Dual fiber paths (one underground, one aerial).
• Result: 99.9% uptime for internet services.

5. Data Center Topologies

How servers, storage, and networks are physically/logically connected:

A. Physical Topologies

B. Network Topologies

Example: Ncell’s DC uses a 3-tier topology:

1. Core Layer: Cisco Nexus 9000 (handles 100Gbps traffic).
2. Aggregation Layer: Juniper MX (load balancing).
3. Access Layer: HP 1920 switches (connects to racks).

6. Disaster Recovery and Business Continuity

Data centers must plan for failures using:

Worked Example: NMB Bank’s Disaster Plan

• Primary DC: Kathmandu (Tier 3).
• Backup DC: Chitwan (Tier 2, 300km away).
• RTO (Recovery Time Objective): 15 minutes.
• RPO (Recovery Point Objective): 5 minutes (data loss ≤5 mins).

7. Emerging Trends in Data Center Design

In the Real World

1. eSewa’s Data Center

   • Idea Used: Hot-aisle/cold-aisle containment + Tier 3 redundancy.
   • How It Works:
     • Cooling: Air flows only in one direction (hot air exits at the top, cold air enters at the bottom), reducing energy use by 30%.
     • Redundancy: Dual UPS + diesel generator ensures zero downtime during load-shedding.
   • Impact: Processes 50,000 transactions/hour without failures.

2. Ncell’s 5G Core Data Center

   • Idea Used: Tier 4-like redundancy (despite being Tier 3) for 99.99% uptime.
   • How It Works:
     • Dual active-active ISPs (NTC + WorldLink).
     • Automated failover for servers (if one node crashes, traffic shifts instantly).
   • Impact: Supports 10M+ 4G/5G users without call drops.

3. Daraz’s Micro-Data Center in Pokhara

   • Idea Used: Edge computing to reduce latency.
   • How It Works:
     • Local VMs (via VxRail) store inventory/data for Pokhara users.
     • Cloud bursting to AWS during Black Friday sales.
   • Impact: 30% faster order processing vs. Kathmandu-based DC.

Exam Tip

1. Memorize the 4-tier classification and PUE formula. Examiners often ask for real-world PUE examples (e.g., Google’s 1.1 vs. Nepali DCs’ 2.0).
2. Compare cooling methods in a table (as above). Expect short-answer questions on hot-aisle vs. liquid cooling.
3. Link Nepali examples to concepts:
   • Ncell’s Tier 3 DC → Redundancy.
   • eSewa’s UPS → Power backup.
   • NTC’s fiber challenges → Network topology.
4. Disaster recovery is a high-weight topic. Know:
   • RTO vs. RPO.
   • Hot vs. cold sites.
   • Nepal Rastra Bank’s Chitwan backup DC.
5. Diagrams are worth marks! Practice drawing:
   • Tiered architecture.
   • Hot-aisle/cold-aisle layout.
   • 3-tier network topology.