Windows Home Server on HP 15s — Part 5: Cost Analysis, Performance Audit, and Hybrid Architecture

This is the engineering case that justifies everything in the previous four parts. We will do the numbers rigorously — not optimistically — and understand exactly where your repurposed laptop stands against commercial alternatives.

1. Electricity Cost: The Real Operating Expense

Your HP 15s-du2077TU's primary ongoing cost as a server is electricity. Unlike a cloud VPS, there are no monthly subscription fees. Let us calculate the exact cost.

1.1 Power Draw Measurements

The Intel Core i5-1035G1 has a configurable TDP of 15W (base) to 25W (boost). In practice, a laptop running as a server draws from the wall:

Operating State	Wall Power Draw
Idle (all Docker services, no AI inference)	~10–14 W
Light load (Nextcloud sync, n8n workflows)	~15–20 W
Medium load (CPU LLM inference, phi3.5)	~22–28 W
Heavy load (8B model inference + other tasks)	~28–35 W
Display on (additional)	+3–5 W
Weighted average (server mode, lid closed)	~13–15 W

For our calculations, we use a conservative average of 14 W. This accounts for the inference periods, idle background services, and periodic Windows Update activity.

1.2 Monthly Energy Consumption

Formula: Energy (kWh) = Power (W) × Hours ÷ 1000

Monthly consumption:
  14 W × 24 hours × 30 days ÷ 1000 = 10.08 kWh/month

Annual consumption:
  10.08 × 12 = 120.96 kWh/year

(For reference: an average Indian household consumes 90–200 kWh/month)
Your server adds approximately 10 kWh to your monthly bill.

1.3 State-Wise Monthly Cost in India (2026)

India uses a telescopic slab pricing system — the rate you pay per unit depends on your total household consumption. If your household already uses 300+ units/month, your server's 10 units will be charged at the highest slab rate.

State	Rate at 200+ units/month	Server Cost/Month	Server Cost/Year
Maharashtra	₹8.50–₹12.83/unit	₹85–₹129	₹1,020–₹1,548
Delhi	₹5.00–₹8.00/unit	₹50–₹80	₹600–₹960
Tamil Nadu	₹5.60–₹11.05/unit	₹56–₹111	₹672–₹1,332
Karnataka	₹5.45–₹8.10/unit	₹55–₹81	₹660–₹972
Uttar Pradesh	₹5.50–₹8.00/unit	₹55–₹80	₹660–₹960
Gujarat	₹4.45–₹7.30/unit	₹45–₹73	₹540–₹876
West Bengal	₹5.49–₹8.50/unit	₹55–₹85	₹660–₹1,020
Average	~₹7/unit	~₹70/month	~₹840/year

Key insight: At the Indian average of ₹7/unit, running the HP 15s as a 24/7 server costs approximately ₹70 per month in electricity.

This is less than a single cup of coffee at Starbucks.

2. Three-Year Total Cost of Ownership (TCO) Analysis

TCO includes: hardware acquisition (amortized), electricity, and any software/service costs.

2.1 HP 15s-du2077TU TCO Breakdown

One-time Hardware Costs (already owned):
  Original laptop price (2020):        ₹48,000
  16 GB RAM upgrade (8 GB × 2):        ₹4,000  (approximate current market price)
  Vertical laptop stand:               ₹800
  Surge protector/UPS (recommended):   ₹2,500
  ──────────────────────────────────────────────
  Total Hardware Investment:           ₹55,300

If the laptop was already owned (bought for personal use):
  Marginal cost of converting to server: ₹7,300 (RAM + stand + UPS)

3-Year Ongoing Costs:
  Electricity (₹70/month × 36 months): ₹2,520
  Domain name (if not owned):           ₹600–₹2,000/year × 3 = ₹6,000
  Backblaze B2 storage (~5 GB):        ₹0 (free tier)
  Cloudflare Tunnel:                   ₹0 (free)
  Software licenses:                   ₹0 (all open-source)
  ──────────────────────────────────────────────
  Total 3-Year Operating Cost:         ₹8,520–₹11,520

3-Year Total (hardware already owned):
  ₹7,300 (RAM + accessories) + ₹8,520 (operating) = ₹15,820

3-Year Total (hardware freshly bought):
  ₹55,300 + ₹8,520 = ₹63,820

2.2 Commercial VPS Comparison: What the Same Money Buys

Let us compare three tiers that provide roughly equivalent capability to what our home server delivers:

Tier 1: Entry Level VPS (basic self-hosting without AI)

Provider	Plan	Monthly Cost	3-Year Cost	RAM	Storage
Hetzner Cloud CPX11	3 vCPU, 4 GB	€4.85 (~₹450)	₹16,200	4 GB	40 GB SSD
Hostinger KVM1	1 vCPU, 4 GB	₹720/mo	₹25,920	4 GB	50 GB NVMe
AWS Lightsail	2 vCPU, 4 GB	$18 (~₹1,500)	₹54,000	4 GB	80 GB SSD
HP 15s server	4C/8T, 16 GB	₹70/mo	₹2,520	16 GB	1.25 TB

At entry level, the HP 15s dominates on every dimension.

Tier 2: Mid-Range VPS (capable of AI inference, but not local)

Note: VPS providers cannot run local LLM inference efficiently because they do not provide GPU access at this price range. If you want AI inference on a VPS, you need either GPU instances (expensive) or you use cloud APIs (more expensive).

Provider	Plan	Monthly Cost	3-Year Cost
Hetzner CPX41	8 vCPU, 16 GB	€14.63 (~₹1,350)	₹48,600
Hostinger KVM4	4 vCPU, 16 GB	₹3,200/mo	₹1,15,200
AWS Lightsail 16GB	4 vCPU, 16 GB	$80 (~₹6,700)	₹2,41,200
HP 15s server	4C/8T, 16 GB	₹70/mo	₹2,520

For a 16 GB equivalent VPS, the 3-year cost difference is staggering.

Tier 3: AI-Capable Cloud (with actual LLM inference)

If you wanted to replicate the local AI capability (running phi3.5 or 8B models) on cloud infrastructure, you would need GPU instances:

Provider	GPU Plan	Monthly Cost	3-Year Cost
AWS G4dn.xlarge (T4 GPU)	4 vCPU, 16 GB, T4 16GB	~~$378 (~~₹31,500)	₹11,34,000
RunPod (community GPU)	RTX 3090 on demand	~₹2,500–₹8,000	₹90,000–₹2,88,000
Groq API (cloud LLM)	~1M tokens/month	~~$25 (~~₹2,100)	₹75,600
HP 15s (local)	Real local GPU (iGPU)	₹70/mo	₹2,520

Local inference wins on cost. The tradeoff is speed (3–15 t/s vs. 100+ t/s on a dedicated GPU), but for personal use, that is acceptable.

2.3 3-Year Savings Summary

Comparison: HP 15s Home Server vs. Hetzner CPX41 (closest equivalent)

Hetzner CPX41 (3 years):               ₹48,600
HP 15s operating cost (3 years):        ₹2,520
────────────────────────────────────────────────
Savings vs. Hetzner:                   ₹46,080

Comparison: HP 15s Home Server vs. AWS Lightsail 16GB (3 years)

AWS Lightsail 16 GB (3 years):         ₹2,41,200
HP 15s operating cost (3 years):         ₹2,520
────────────────────────────────────────────────
Savings vs. AWS:                       ₹2,38,680 (≈ ₹2.4 Lakh)

Including hardware investment (RAM + accessories):
  Net savings vs. Hetzner:             ₹46,080 - ₹7,300 = ₹38,780
  Net savings vs. AWS:                 ₹2,38,680 - ₹7,300 = ₹2,31,380

The payback period for the RAM upgrade investment (₹4,000) vs. the cheapest comparable VPS (Hetzner ₹1,350/month) is less than 4 months.

3. Performance Audit: WSL2 vs. Bare-Metal Linux

For readers who want to know exactly what they sacrifice by staying on Windows instead of installing bare-metal Ubuntu Server:

3.1 CPU Performance

Benchmark	WSL2 (Windows 10)	Bare-Metal Ubuntu	Difference
sysbench CPU (events/s)	~12,400	~12,700	-2.4%
Node.js compute	~96% of native	100%	-4%
Python compute	~97% of native	100%	-3%
Docker image build	~94% of native	100%	-6%
LLM token speed	Identical	Identical	0% (memory-bound)

CPU overhead is negligible for server workloads. The ~3–5% difference is imperceptible in production use.

3.2 Memory Overhead

Metric	WSL2	Bare-Metal	Impact
Base OS overhead	~2.8 GB (Windows)	~180 MB (Ubuntu)	-2.6 GB usable
Effective container RAM	~13 GB (of 16)	~15.8 GB (of 16)	-2.8 GB
RAM for AI models	Up to ~10 GB	Up to ~13 GB	Can't run 13B

The 2.8 GB Windows overhead is the most significant real trade-off. On 16 GB, you can still run 7B–8B models comfortably. The 13B size class is borderline (possible with model unloading everything else, risky) vs. comfortable on bare metal.

3.3 Disk I/O Performance

Access Pattern	WSL2	Bare-Metal	Difference
Internal VHDX sequential read	~1,400 MB/s	~1,500 MB/s	-7%
Internal VHDX random 4K read	~190K IOPS	~200K IOPS	-5%
Cross-OS /mnt/c sequential read	~120 MB/s	N/A	-92% vs NVMe
Docker build (volumes in VHDX)	Near-native	Native	-5%

The critical rule bears repeating: Never store active Docker volumes on /mnt/c/ or any cross-OS mount. Always store them inside the WSL2 VHDX (~/server/). This is the single most impactful performance decision.

3.4 WSL2 Latency Characteristics

Operation	WSL2 Latency	Bare-Metal Latency	Notes
Container-to-container HTTP	~0.2 ms	~0.1 ms	Both on same network
Container-to-Windows host HTTP	~1–2 ms	N/A	WSL2 NAT bridge
Disk fsync (database writes)	~0.5–1 ms	~0.3–0.8 ms	Acceptable for SQLite
systemctl service start	~300 ms	~150 ms	systemd overhead

For web application response times (where your services add 5–50 ms of business logic anyway), these sub-millisecond differences are invisible.

3.5 The Verdict: When to Choose Which

Stay on Windows + WSL2 if:

You use the laptop as a personal computer and need Windows for specific applications (Office, software that requires Windows)
You prefer not to reinstall the OS
Your workload fits comfortably in 13 GB of container RAM
You do not need models larger than 8B parameters

Switch to bare-metal Ubuntu Server if:

This laptop is 100% dedicated as a server (never used as a desktop)
You want maximum container density and RAM headroom for 13B+ models
You prefer a simpler architecture (no WSL2 layer, no Windows quirks)
You are comfortable with an SSH-only workflow

4. 15-Dimension Evaluation Matrix

A systematic comparison of the self-hosted approach vs. cloud VPS:

Dimension	HP 15s (WSL2)	Hetzner VPS	AWS Lightsail
Monthly cost	₹70	₹1,350	₹6,700
RAM available	~13 GB	16 GB	16 GB
Storage	1.25 TB	40 GB	80 GB
LLM inference	✅ Yes (CPU, local)	❌ No GPU	❌ No GPU
Network speed	LAN: 1 Gbps / WAN: ISP speed	1 Gbps	1 Gbps
Uptime guarantee	Depends on power/ISP	99.9% SLA	99.9% SLA
Data privacy	✅ Your hardware	⚠️ EU jurisdiction	⚠️ US company
Backup control	✅ Full control	✅ Backups available	✅ Snapshots
Physical security	Your premises	Data center	Data center
Setup complexity	High (this guide!)	Low-Medium	Low
Free tier possible	✅ Yes (just power)	❌ No	❌ No
Data egress cost	₹0	₹0	₹8.50/GB
Scalability	Fixed hardware	Resize with click	Resize with click
ISP dependency	High (CGNAT risk)	None (datacenter)	None (datacenter)
Energy efficiency	14 W	~30 W/4 vCPU	Shared, unknown

Net assessment: For individuals in India who already own compatible hardware and can tolerate the setup complexity, self-hosting on the HP 15s is dramatically more cost-effective, more private, and surprisingly more capable (local AI) than any commercial VPS at equivalent price points.

5. Hybrid Architecture: Best of Both Worlds

Pure self-hosting has one real weakness: availability during power outages, ISP downtime, or laptop failure. The solution is a hybrid architecture that uses free cloud services to cover the gaps.

┌─────────────────────────────────────────────────────────────────────┐
│                    HYBRID ARCHITECTURE OVERVIEW                      │
│                                                                      │
│  PUBLIC INTERNET                                                     │
│       │                                                              │
│       ▼                                                              │
│  ┌─────────────────────────────────────────────────────────────────┐│
│  │  CLOUDFLARE EDGE (Global CDN — Free)                            ││
│  │  ├── DNS (authoritative, global)                                ││
│  │  ├── WAF (Web Application Firewall — Free tier)                ││
│  │  ├── CDN Cache (static assets cached at edge)                  ││
│  │  ├── Pages (blog.yourdomain.com → Cloudflare Pages, free)      ││
│  │  └── Tunnel (dynamic apps → HP 15s home server)                ││
│  └────────────────────────────────────┬────────────────────────────┘│
│                                       │                              │
│                  ┌────────────────────┼────────────────────┐        │
│                  │                    │                    │        │
│                  ▼                    ▼                    ▼        │
│         ┌─────────────┐    ┌──────────────────┐  ┌─────────────┐  │
│         │ STATIC TIER │    │  DYNAMIC TIER    │  │  DATA TIER  │  │
│         │ (Cloud Free)│    │  (HP 15s Local)  │  │ (Hybrid)    │  │
│         │             │    │                  │  │             │  │
│         │ Cloudflare  │    │ Nextcloud        │  │ Neon DB     │  │
│         │ Pages       │    │ Vaultwarden      │  │ (Postgres   │  │
│         │ (blog,      │    │ n8n workflows    │  │  serverless │  │
│         │  portfolio) │    │ Odysseus agents  │  │  free tier) │  │
│         │             │    │ Ollama (AI)      │  │             │  │
│         │ GitHub      │    │ Open WebUI       │  │ Backblaze   │  │
│         │ (code,      │    │ Flowise          │  │ B2 (backups │  │
│         │  CI/CD)     │    │ Uptime Kuma      │  │  ~free 10GB)│  │
│         └─────────────┘    └──────────────────┘  └─────────────┘  │
└─────────────────────────────────────────────────────────────────────┘

5.1 Tier 1: Static Assets → Cloudflare Pages (Free)

Your blog, portfolio, documentation site, and any static frontend assets should be hosted on Cloudflare Pages — not on your home server. Reasons:

Cloudflare Pages is free (unlimited static asset bandwidth).
It is globally distributed across 300+ PoPs — ~20 ms latency anywhere.
If your home server is offline, static pages remain available.
Zero configuration SSL/TLS.

Deployment:

# In your blog repository (e.g., blog.oriz.in)
# .github/workflows/deploy.yml

name: Deploy to Cloudflare Pages
on:
  push:
    branches: [main]

jobs:
  deploy:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0

      - name: Install Node.js
        uses: actions/setup-node@v4
        with:
          node-version: '20'

      - name: Install dependencies and build
        run: npm ci && npm run build

      - name: Publish to Cloudflare Pages
        uses: cloudflare/pages-action@v1
        with:
          apiToken: ${{ secrets.CLOUDFLARE_API_TOKEN }}
          accountId: ${{ secrets.CLOUDFLARE_ACCOUNT_ID }}
          projectName: blog-oriz-in
          directory: ./dist
          gitHubToken: ${{ secrets.GITHUB_TOKEN }}

5.2 Tier 2: Dynamic Apps → Home Server (via Cloudflare Tunnel)

Heavy, stateful applications belong on your home server:

Nextcloud — 1 TB of local storage impossible on free cloud tiers.
Ollama + Open WebUI — no cloud provider gives you local inference.
Vaultwarden — your password vault must be under your control.
n8n — complex workflows requiring persistent state and file access.
Odysseus — agent workspace requiring local model and file access.

5.3 Tier 3: Serverless Database → Neon DB (Free Tier)

For applications that need a PostgreSQL database but where you do not want to manage the DB server yourself (e.g., a side project API):

Neon DB Free Tier (2026):

0.5 GB storage
190 compute hours/month
Automatic scaling to zero (no idle cost)
PostgreSQL 16 compatible
Connection string format compatible with any pg client

# Use Neon for API databases while keeping Nextcloud/n8n local
# Example: a simple link shortener API on Cloudflare Workers + Neon

# In your Cloudflare Worker:
# DATABASE_URL = "postgresql://user:[email protected]/neondb?sslmode=require"

5.4 Failure Mode Analysis: What Fails When

Understanding what breaks under each failure scenario is essential for planning:

Failure Scenario	Static Sites	Dynamic Apps	AI Features	Recovery Time
Home Internet goes down	✅ Still up	❌ Down	❌ Down	ISP SLA (usually 1–4 hrs)
BSNL/Jio CGNAT issue	✅ Still up	❌ Tunnel down	❌ Down	Fix: re-auth tunnel
Windows Update restart	✅ Still up	~5 min down	~5 min down	Autologon restores
Power outage (no UPS)	✅ Still up	❌ Hard crash	❌ Down	Need AC + Autologon
Laptop hardware failure	✅ Still up	❌ Down	❌ Down	Buy new hardware
Cloudflare outage	⚠️ May be affected	❌ Tunnel down	✅ LAN works	~rare, short
SSD failure	✅ Still up	❌ Data at risk	❌ Down	Restore from Restic

Mitigation strategies:

Power outages: A ₹2,500 mini-UPS (like APC Back-UPS BX600C) provides 10–15 minutes of runtime at 14W — enough to survive 99% of Indian power cuts.
Hardware failure: Keep a weekly Restic backup to B2. If the laptop dies, restore to any machine with WSL2 in under 2 hours.
ISP downtime: Secondary mobile hotspot (Jio/Vi/Airtel) + cloudflared can failover. Or accept that personal services go down when internet does — for home use, this is usually fine.

5.5 Monthly Cost Summary: Full Hybrid Stack

Complete Hybrid Architecture — Monthly Operating Cost (India):

  HP 15s electricity:          ₹70
  Domain name (amortized):     ₹50 (₹600/year)
  Cloudflare Tunnel:           ₹0 (free)
  Cloudflare Pages:            ₹0 (free)
  GitHub (private repos):      ₹0 (free with up to 2,000 CI mins)
  Neon DB (free tier):         ₹0 (free up to 0.5 GB)
  Backblaze B2 (up to 10 GB): ₹0 (free)
  ────────────────────────────────────
  TOTAL:                       ₹120/month

For comparison:
  A comparable Hetzner + Cloudflare + managed DB stack: ₹3,500–₹5,000/month
  A comparable AWS stack:                               ₹8,000–₹15,000/month

Annual savings vs. Hetzner equivalent:  ₹40,560–₹58,560
Annual savings vs. AWS equivalent:      ₹93,360–₹1,76,160

6. The Final Complete Service Inventory

Here is everything you now run on a ₹70/month home server:

┌─────────────────────────────────────────────────────────────────────┐
│          COMPLETE HP 15s HOME SERVER — SERVICE INVENTORY             │
├─────────────────────────────────┬───────────────────────────────────┤
│ Service                         │ Cloud Equivalent (Monthly cost)   │
├─────────────────────────────────┼───────────────────────────────────┤
│ Nextcloud (cloud storage)       │ Google One 100 GB: ₹170/mo       │
│ Vaultwarden (password manager)  │ 1Password Family: ₹450/mo        │
│ n8n (workflow automation)       │ Make.com Starter: ₹1,900/mo      │
│ Open WebUI + Ollama (AI chat)   │ ChatGPT Plus: ₹1,700/mo         │
│ Odysseus (AI workspace)         │ No direct equivalent; Claude Pro: ₹1,700/mo |
│ Flowise (AI pipeline builder)   │ LangSmith Plus: ₹2,500/mo       │
│ Uptime Kuma (monitoring)        │ Better Uptime Basic: ₹400/mo    │
│ Portainer (container mgmt)      │ Docker Hub Pro: ₹700/mo         │
│ Qdrant (vector database)        │ Qdrant Cloud Starter: ₹1,700/mo │
│ CloudFlare Tunnel               │ Cloudflare Zero Trust: ₹0        │
├─────────────────────────────────┼───────────────────────────────────┤
│ Combined Cloud Equivalent:      │ ₹11,220/month                    │
│ Your Cost:                      │ ₹70/month (electricity only)     │
│ Monthly Savings:                │ ₹11,150/month                    │
│ Annual Savings:                 │ ₹1,33,800/year                   │
└─────────────────────────────────┴───────────────────────────────────┘

7. Roadmap: What to Add Next

This guide covers the foundation. As you grow comfortable with the stack, here are logical next steps:

7.1 Immich — Self-Hosted Google Photos

If you use Google Photos, Immich replaces it with local storage + machine learning for face recognition and object search. With your 1 TB HDD, you can store years of photos.

# Add to docker-compose.yml
immich_server:
  image: ghcr.io/immich-app/immich-server:release
  container_name: immich_server
  volumes:
    - /mnt/d/server-data/photos:/usr/src/app/upload
  ports:
    - "127.0.0.1:2283:2283"
  environment:
    DB_HOSTNAME: immich_postgres
    DB_USERNAME: immich
    DB_PASSWORD: immich_password
    DB_DATABASE_NAME: immich
  depends_on:
    - immich_postgres
    - immich_redis
  networks:
    - server_net

7.2 Jellyfin — Self-Hosted Streaming

Jellyfin streams your local movie and TV collection with full library management, subtitle support, and remote access through Cloudflare Tunnel. The i5-1035G1 supports Intel Quick Sync Video for hardware-accelerated transcoding — though this requires passing the GPU through to the Docker container (complex on WSL2; easier on bare-metal Linux).

7.3 Gotify — Self-Hosted Push Notifications

Replace Telegram bots for alerts with Gotify — a lightweight notification server with Android/iOS apps. All health alerts, backup completions, and system events route through your own notification server.

7.4 Gitea / Forgejo — Self-Hosted Git

For code that is too sensitive for GitHub (internal tooling, scripts with API keys, private work), Gitea provides a full GitHub-like experience: repositories, issues, CI/CD via Gitea Actions, and web-based code editing.

7.5 Tailscale — Private WireGuard Mesh Network

Cloudflare Tunnel gives you public HTTPS access. Tailscale gives you a private VPN mesh that allows your phone, other computers, and devices to access the server directly via private IP — bypassing Cloudflare entirely. Free for up to 100 devices.

# Install Tailscale on Windows
winget install --id Tailscale.Tailscale -e --silent
# Follow the login prompt
# Your HP 15s gets a stable private IP like 100.x.x.x
# Access Portainer at http://100.x.x.x:9000 from your phone

Final Summary: Series Complete

Over five parts, we have:

Part	What We Built
1	Foundation: Windows power config, WSL2, native Docker Engine
2	Networking: Cloudflare Tunnel, Caddy proxy, core services
3	AI Stack: Ollama + Open WebUI + Odysseus + Flowise + Qdrant
4	Operations: Self-healing monitoring, Telegram alerts, Restic backup
5	Analysis: Electricity cost, 3-year TCO, hybrid architecture

Your HP 15s-du2077TU, purchased in 2020 for everyday tasks, now runs the equivalent of ₹11,220/month worth of cloud services — for ₹70/month in electricity.

The laptop that was gathering dust is now a private cloud, an AI research station, an automation engine, a password vault, and a file server — all under your control, in your home, on hardware you already own.

This series was written for the HP 15s-du2077TU (i5-1035G1, 16 GB DDR4-2666, 256 GB NVMe + 1 TB HDD, Windows 10 Home). All scripts and configurations are tested against this exact hardware profile.