Slashing data transfer costs in AWS

Original Article ↗ Hacker News Discussion ↗

Engineers dissect ways to slash Amazon Web Services (AWS) data transfer costs, from routing cross–availability zone traffic through S3 or CloudFront to exploiting Lightsail and ECR allowances—while warning that some of these tactics skirt AWS terms and could be shut down. Many argue that AWS egress pricing is vastly above underlying network costs and functions as a lock-in mechanism, prompting comparisons with far cheaper VPS and bare-metal providers or Cloudflare R2 for bandwidth-heavy workloads. Others counter that managed cloud services, discounts at scale, and reduced operational burden still justify AWS for many organizations, stressing that the right choice depends on workload patterns, reliability needs, and in-house expertise.

AWS data transfer cost hacks

  • Main trick discussed: route inter‑AZ traffic through S3 with VPC endpoints, exploiting free uploads and free in‑region S3 reads to slash cross‑AZ charges.
  • Others mention:
    • Using CloudFront’s free egress tier (1 TB/month) as an out‑of‑AWS escape hatch.
    • Lightsail instances as bandwidth “proxies” because they include large transfer quotas.
    • Public ECR images (with encrypted payloads) as a cheap way to push limited data out.
  • People note S3 “transient” storage often appears to cost ~$0 if objects are short‑lived; some speculate about coarse billing granularity or sampling.

Are these loopholes and will AWS close them?

  • Some call this tax‑avoidance‑style “loophole” use of AWS’s pricing, expecting AWS to react once usage is large enough.
  • Others argue this is by design: S3 and CloudFront are deliberately cheap in specific patterns to encourage those architectures.
  • GCP is cited as having already closed a similar cross‑region storage loophole; several say AWS could change S3 billing to kill this pattern.
  • Lightsail ToS explicitly forbid using it to dodge other AWS data fees; enforcement is seen as rare but possible, especially for heavy abusers.

Cloud bandwidth pricing and economics

  • Strong consensus that raw bandwidth is extremely cheap at the wholesale level; AWS/GCP egress is seen as a very high‑margin line item and a lock‑in tool (free ingress, expensive egress).
  • Some argue pricing also shapes traffic patterns and capacity planning, not just profit.
  • Others counter that smaller VPS/bare‑metal providers offer much cheaper or flat‑rate bandwidth while remaining profitable, so hyperscaler egress pricing looks like pure markup.

Cloud vs VPS / bare metal / on‑prem

  • One camp: if you’re doing “sophisticated cloud‑cost analysis,” you might be better off with VPS or colo where bandwidth is cheap and pricing is simpler.
  • Counter‑camp: running serious on‑prem or colo reliably is hard and labor‑intensive; salaries, on‑call, DR, security, and governance often outweigh AWS costs.
  • Multiple anecdotes:
    • Huge savings moving from AWS to providers like Hetzner (e.g., $50k→$800/month), with acceptable reliability.
    • A failed colo migration where a single sysadmin became a point of failure; outage forced a return to AWS.

When cloud makes sense; when it doesn’t

  • Cloud praised for elasticity, managed services, and fast self‑service (vs. ticket‑driven on‑prem bureaucracy).
  • Others stress it’s just a tool: for stable, bandwidth‑heavy workloads with in‑house expertise, VPS/colo can be 5–10× cheaper.
  • Many agree cost analysis is essential either way; “cloud good” vs “cloud bad” mantras are seen as unhelpful compared to explicit trade‑off analysis.