papers_cost_comparison
Scientific Papers Storage — Cost Comparison & Analysis
Scenario 1: MVP (100K Papers, 300GB)
AWS S3 + RDS + Qdrant (Recommended for Prototype)
Monthly Recurring Costs:
┌─────────────────────────────────────────┐
│ Component Cost Notes │
├─────────────────────────────────────────┤
│ S3 Storage (300GB) $7 0.023/GB │
│ S3 Requests $40 100K/mo │
│ RDS PostgreSQL $40 t3.med │
│ Qdrant (t3.large) $60 no HA │
│ NAT Gateway $45 egress │
│ ─────────────────────────────────────── │
│ SUBTOTAL $192/mo │
│ 6-month pilot $1,150 total │
│ 12-month $2,304 │
└─────────────────────────────────────────┘
Cost per Paper: $0.0019/month
Breakdown:
- Storage dominates at scale (not MVP)
- Requests + networking: $85/mo (44%)
- Compute (DB + search): $100/mo (52%)
- Highly elastic — drops with reduced queries
Alternative: MinIO Self-Hosted (100K papers)
One-Time Costs (Year 1):
┌──────────────────────────────┐
│ Hardware $2,500 │
│ • 8×4TB HDD $800 │
│ • 1U server + network $1,700 │
│ │
│ MinIO License $24,000 │
│ (Standard tier, 1yr) │
│ │
│ Setup & Ops $2,000 │
│ ──────────────────────────────│
│ TOTAL YEAR 1 $28,500 │
└──────────────────────────────┘
Monthly (Recurring After Year 1):
┌──────────────────────────────┐
│ License Renewal $2,000 │
│ Power + Cooling $300 │
│ Network (1Gbps) $200 │
│ Personnel (part-time) $1,500 │
│ ──────────────────────────────│
│ TOTAL MONTH 2+ $4,000 │
│ (or $20K/year) │
└──────────────────────────────┘
Breakeven Analysis:
AWS cumulative cost over time:
Year 1: $2,304
Year 2: $4,608
Year 3: $6,912 ← MinIO pays for itself here
Year 4: $9,216
Year 5: $11,520
MinIO cumulative cost over time:
Year 1: $28,500
Year 2: $48,500 ← Breakeven at ~18 months vs AWS
Year 3: $68,500
Year 4: $88,500
Year 5: $108,500
VERDICT: AWS better for <18 months, MinIO better for 5+ year operation
Scenario 2: Growth (1M Papers, 3TB)
AWS S3 Intelligent-Tiering (Recommended for Production)
Monthly:
┌─────────────────────────────────────────┐
│ Component Cost Notes │
├─────────────────────────────────────────┤
│ S3 Storage (3TB) $75 intelligent │
│ S3 Requests $50 1M/mo │
│ RDS PostgreSQL $200 r6i.xlg │
│ Qdrant 3-node $450 HA cluster│
│ NAT Gateway $75 egress │
│ ─────────────────────────────────────── │
│ SUBTOTAL $850/mo │
│ 12-month $10,200 │
│ 3-year $30,600 │
└─────────────────────────────────────────┘
Cost per Paper: $0.00085/month
MinIO Self-Hosted (1M Papers, 3TB Usable)
Infrastructure (with 1.5x redundancy for reliability):
┌──────────────────────────────────┐
│ Storage Needed (1.5x RS): 4.5TB │
│ • 6×4TB HDD drives $1,200 │
│ • 2×1U servers $4,000 │
│ • Network (redundant) $1,000 │
│ ─────────────────────────────────│
│ Hardware Total $6,200 │
│ │
│ MinIO Enterprise License $24,000 │
│ (1 year, larger deployment) │
│ ─────────────────────────────────│
│ Initial Setup $2,000 │
│ ─────────────────────────────────│
│ TOTAL YEAR 1 $32,200 │
└──────────────────────────────────┘
Monthly Ops (Years 2+):
License: $2,000
Power + Cooling: $500
Network: $300
Personnel (1 FTE): $4,000
─────────────────────────
TOTAL/month: $6,800
TOTAL/year: $81,600
Comparison Table (1M Papers):
| Timeframe | AWS S3 | MinIO | Winner |
|---|---|---|---|
| Year 1 | $10,200 | $32,200 | AWS (+$22K) | |
| Year 2 | $20,400 | $113,800 | AWS (+$93K) | |
| Year 3 | $30,600 | $195,400 | AWS (+$165K) | |
| Year 5 | $51,000 | $439,000 | AWS (+$388K) |
Note: MinIO assumes self-hosting with dedicated ops team. If outsourcing ops (managed MinIO on cloud), costs rise to $200–400/mo, eliminating savings.
Scenario 3: Enterprise (10M+ Papers)
AWS with Reserved Instances (RI) Discount
Monthly with 3-year RI (30% discount):
┌────────────────────────────┐
│ S3 Storage (30TB) $750 │
│ S3 Requests $200 │
│ RDS (multi-region) $1,200 │
│ Qdrant 10-node $2,000 │
│ ────────────────────────── │
│ TOTAL $4,150 │
│ Annual $49,800 │
│ 3-year $149,400 │
└────────────────────────────┘
Cost per Paper: $0.00050/month (1/2 of growth scenario)
MinIO Managed (by Third-Party)
Example: Stackhero MinIO Managed Service (1PB)
Stackhero MinIO Pricing:
┌────────────────────────────┐
│ 1PB Storage N/A│
│ (actually limited to 1TB) │
│ │
│ 1TB (managed) $500/mo│
│ ────────────────────────── │
│ TOTAL/year $6,000 │
│ 3-year $18,000 │
└────────────────────────────┘
For 10M papers (30TB), need multiple instances = uneconomical
Data Transfer Costs (Often Overlooked)
AWS Egress Charges (Major Factor)
Scenario: 100K papers, 30% accessed monthly (avg 2 downloads/access)
Monthly Downloads: 100K × 30% × 2 = 60,000 downloads
Avg File Size: 3MB
Total egress: 60,000 × 3MB = 180GB
Egress cost: 180GB × ⟨MATH:4⟩16.20/month
Annual egress: ~$194
NOTE: First 100GB/month FREE, so actual cost starts at month 2:
Months 1-12: $0 (under free tier)
Month 13+: $16.20/month
For higher-traffic scenarios (30% daily access):
Monthly: 100K × 30% × 2 × 30 days = 1.8M downloads
Total egress: 1.8M × 3MB = 5.4TB
Cost: (5.4TB - 100GB) × ⟨MATH:5⟩486/month
Annual: $5,832 (becomes major cost driver)
SOLUTION: Add CloudFront CDN
CDN cache hit rate: 80% → reduces egress to $97/month
Recommendation Matrix
Choose AWS S3 If:
- Prototype phase (< 6 months)
- Unpredictable access patterns
- Need multi-region replication
- Team inexperienced with ops
- Budget flexibility (pay-as-you-go)
- < 5 years planned operation
Cost Profile: Linear growth with scale; no upfront capex
Choose MinIO If:
- 5+ year operation planned
- Predictable, high-volume access
- No egress charges acceptable (cost savings)
- Data residency / sovereignty required
- Dedicated ops team available
- > 20TB data, growing steadily
Cost Profile: High upfront capex; breakeven at 18–24 months; superior long-term economics
Choose Yandex Object Storage If:
- Operating in Russia / EU
- GDPR compliance required
- Want AWS-like experience without US data
- Cost similar to AWS (maybe 15–20% cheaper)
Note: Pricing not fully disclosed; estimate based on CloudFlare R2 comparables (~$0.015/GB storage)
Hidden Costs (Always Budget For)
┌──────────────────────────────────────────┐
│ Often Forgotten... Typical Cost│
├──────────────────────────────────────────┤
│ LLM API calls (enrichment) ~$100-200/mo │
│ Tesseract OCR (scanned PDFs) ~$50-100/mo │
│ Monitoring (DataDog/New Relic) ~$50/mo │
│ Backup storage (Glacier) ~$20-50/mo │
│ CDN (CloudFront) ~$10-50/mo │
│ Custom domain / SSL ~$10-20/mo │
│ ─────────────────────────────────────────│
│ TOTAL "HIDDEN" ~$300-450/mo │
│ (Can double your base cost!) │
└──────────────────────────────────────────┘
Summary Recommendation
For Макс (Manager Decision):
Phase 1 (Months 0–6): AWS S3 Proof-of-Concept
- Upfront investment: $0 capex
- Monthly burn: ~$192
- 6-month pilot: $1,150
- Decision point: Assess usage patterns, team capabilities, long-term strategy
Phase 2 (Months 6–12): Scale Assessment
- If <5TB and unpredictable: Continue AWS
- If 20TB+ and predictable: Plan MinIO migration
- If GDPR needed: Migrate to Yandex
Phase 3 (Months 12+): Optimization
- AWS: Add CloudFront CDN, use S3 Intelligent-Tiering, implement RI
- MinIO: Optimize hardware utilization, automate ops
- Yandex: Evaluate local S3 API features, cost vs AWS
ROI Calculation for Manager
Assumption: Building this system saves 2 engineers × 20 hrs/week × $50/hr (opportunity cost)
Weekly savings: 2 × 20 × ⟨MATH:6⟩2,000
Annual savings: $104,000
System cost (AWS Year 1): $2,304
Return: ⟨MATH:7⟩2,304 = 45x ROI
Payback period: 8 days (!)
Conclusion: Even at conservative estimates, system ROI is exceptional. Cost is not a blocker.