Egress Costs: Inter-region Pricing, Optimization Strategies

Network egress cost là "silent killer" của cloud bills. Vì không visible (transparent ở layer-3 networking), team không nhận thức chi phí cho tới khi bill tới.

Một cách hiểu sai: "Egress bé tí, không quan trọng". Thực tế: multi-region data synchronization một lần có thể tốn hàng triệu VND.

---

Egress Pricing Structure

Internet Egress (Internet-bound traffic)

Tier 1: 0–1 TB/month       $0.12/GB
Tier 2: 1–10 TB/month      $0.08/GB
Tier 3: 10+ TB/month       $0.06/GB

Example:

100 GB/month internet egress:
  Cost = 100 × $0.12 = $12

10 TB (10,000 GB)/month internet egress:
  Tier 1: 1,000 × $0.12 = $120
  Tier 2: 9,000 × $0.08 = $720
  Total = $840

Inter-Region Egress (Between GCP regions)

Inter-region egress not on same continent: $0.02/GB

Examples:
  us-central1 → us-east1:    $0.02/GB (same continent)
  us-central1 → europe-west1: $0.02/GB (same continent)
  us-central1 → asia-south1:  $0.02/GB (same continent, but premium)
  
Actually, all inter-region egress: $0.02/GB

Same-Region / Same-Zone Traffic

Same region, different zones:   FREE
Same zone (technically):        FREE

This is key for cost optimization!

---

Cost Impact: Real Examples

Example 1: Daily Data Sync (TBs)

Scenario: Sync 100 GB database daily from us-central1 to asia-south1

Daily cost: 100 GB × $0.02 = $2
Monthly cost: $2 × 30 = $60
Annual cost: $60 × 12 = $720

With larger dataset (10 TB daily):
Monthly: 10,000 GB × $0.02 × 30 = $6,000/month = $72,000/year

Example 2: CDN Content Delivery

Scenario: Serve 1 TB/day content from us-central1 to asia-southeast1 users

Without CDN (internet egress):
  1 TB/day × 30 days = 30 TB/month
  30 TB × $0.06 = $1,800/month

With Cloud CDN (origin → cache):
  Only cache misses egress from origin
  Assume 10% miss rate = 3 TB egress
  3 TB × $0.06 = $180/month
  
Savings: $1,620/month (90%)

---

Optimization Strategies

Strategy 1: Colocation (Same Region)

Principle: Data and compute in same region = no inter-region egress.

Bad architecture:
  Database (us-central1) → Application (asia-south1)
  → Inter-region egress for every query

Good architecture:
  Database (asia-south1) → Application (asia-south1)
  → Same-region, no egress cost

Implementation:

• Multi-region data: replicate database to each region
• Application: run in same region as primary data
• Trade-off: Multi-region replication cost vs egress savings

Strategy 2: Cloud CDN (Cache Origin)

CDN mechanics:
  - Edge cache near end-user
  - Cache hit: data from edge, no origin egress
  - Cache miss: data from origin, 1x egress cost

Deployment:
  gcloud compute backend-services update my-backend \
    --enable-cdn \
    --cache-mode=CACHE_ALL_STATIC

Savings potential: 70–90% egress reduction (assuming high hit rate).

Strategy 3: Cloud Interconnect

For bulk, continuous data flow, Cloud Interconnect cheaper than public internet:

Dedicated Interconnect:
  Setup: $0.30/hour per port (10 Gbps)
  Usage: No per-GB egress charge
  
Breakeven:
  $0.30/hour × 730 hours = $219/month minimum
  
If sending > 300 TB/month (breakeven at $0.02/GB × 300 TB = $6k/month):
  Cost via Interconnect: $219/month (fixed)
  Cost via internet: $6k+/month
  Savings: ~$6k/month

Strategy 4: Compression

Every GB compressed = less egress cost.

Scenario: Replicate 500 GB database nightly

Uncompressed:
  500 GB × $0.02 = $10/sync
  
Compressed (2x compression):
  250 GB × $0.02 = $5/sync
  
Compression overhead: 30 seconds CPU
vs Savings: $5/sync × 30 syncs/month = $150/month

ROI: Positive (save money + reduce time)

Strategy 5: Data Locality via BigQuery

Bad:
  Store data in us-central1
  Query from asia-south1
  → Data transfer cost + inter-region egress

Good:
  BigQuery regional dataset (asia-south1)
  Query locally
  → No data transfer, local query, no egress

---

Egress Cost Visibility

Query Cloud Billing for Egress

SELECT
  service.description,
  sku.description,
  SUM(cost) as egress_cost,
  SUM(usage.amount) as usage_gb
FROM `project.billing_dataset.gcp_billing_export_v1`
WHERE sku.description LIKE '%Egress%'
  AND DATE(usage_start_time) >= DATE_SUB(CURRENT_DATE(), INTERVAL 30 DAY)
GROUP BY service.description, sku.description
ORDER BY egress_cost DESC;

VPC Flow Logs for Traffic Analysis

# Enable VPC Flow Logs
gcloud compute networks update default \
  --enable-flow-logs

# Analyze traffic patterns
gcloud logging read \
  "resource.type=gce_instance AND 
   jsonPayload.dest_country='IN'" \
  --limit 100

---

Regional Cost Matrix

From / To       | us-central1 | europe-west1 | asia-south1
us-central1     | FREE        | $0.02        | $0.02
europe-west1    | $0.02       | FREE         | $0.02
asia-south1     | $0.02       | $0.02        | FREE

Insight: All inter-region egress uniform ($0.02/GB), so regional choice driven by:

1. Data residency requirements
2. Latency (GCP backbone helps)
3. User location

---

Anti-Patterns

Anti-Pattern 1: Uncontrolled Multi-Region Replication

Scenario: Replicate entire database to 5 regions daily.

Impact:

Database size: 5 TB
Replication: 4 × 5 TB = 20 TB egress/day
Cost: 20 × $0.02 × 30 = $12,000/month

But use case: Dev/test replication (not needed)

Fix: Selective replication (only critical data, hot regions).

Anti-Pattern 2: No CDN for Public Content

Scenario: Serve static content (images, JS) from origin every time.

Impact:

10 Gbps continuous traffic = ~38 TB/day
Cost: 38 × $0.06 = $2,280/month

With CDN (80% hit rate):
Cost: (38 × 0.2) × $0.06 = $456/month

Fix: Enable Cloud CDN for public content.

Anti-Pattern 3: Ignoring Backup Egress

Scenario: Daily backup to different region, 100 GB/day, but not tracked.

Impact:

Cost: 100 × $0.02 × 30 = $60/month (seems small, but ...)
Annual: $720 + RPO cost + restoration time cost

If using lifecycle policy + local snapshot:
Cost: $0 (same region, free) + storage cost

Fix: Use local snapshots, lifecycle policies for warm storage.

---

Practical Optimization Checklist

☐ Query egress cost in billing (identify top traffic flows)
☐ Enable Cloud CDN for public content
☐ Collocate databases with applications (same region)
☐ Compress large data transfers (2x reduction typical)
☐ Use Interconnect if sustained > 300 TB/month inter-region
☐ Monitor VPC Flow Logs for unexpected egress patterns
☐ Document data flow in architecture (track egress)
☐ Alert on egress cost spikes (Pub/Sub anomaly detection)

Expected savings: 30–60% egress cost reduction

---

References

• Network Pricing
• Cloud CDN Overview
• Cloud Interconnect Pricing
• VPC Flow Logs