Multi-CDN setups still generate origin fill on cache misses — each CDN’s edge transfer and origin-pull path can meter differently. Estimate fill ≈ client_GB × (1−CHR) per CDN lane; keep $/GB Unknown until verified. Use Mode C. Origin fill is not S3 CRR and not Internet DT-out — decode AWS origin talkers via the Cost Explorer decoder when CloudFront / S3 lines appear on the same bill.
Updated
Two CDN logos in front of one origin still share miss-driven fill — map edge vs origin meters per provider before blending.
Multi-CDN is an availability and bargaining pattern — not a single meter. PipeToll models each lane’s origin fill with CHR before comparing logos. See also CloudFront vs Cloudflare (already shipped) and multi-CDN.
Path templates to model
Active-active split — 50/50 clients; two cold caches can double origin pull versus one warm CDN.
Failover-only secondary — secondary CHR near zero during drills; budget drill fill.
Shielded origin — mid-tier reduces origin GB but adds another hop to price.
Object storage origin — miss fill may bill as S3 DT-out or private path; keep Unknown until cited.
Bypass / signed direct — large clients skipping CDN push 100% to origin egress.
Common traps
One blended CDN $/GB — Collapsing edge-to-client and origin-fill into a single “CDN $/GB” hides two meters (and often two vendors). Keep Mode A-style edge lines apart from Mode C fill.
Duplicate fill on dual miss — Ignoring duplicate fill when both CDNs miss the same surge understates origin pull versus one warm lane. Model CHR per logo.
Multi-CDN always cheaper — Assuming a second logo always lowers origin cost invents savings. Two cold caches can raise fill; drill/failover months need their own budget.
Skipping Mode C per lane — Blending both lanes into one Mode C run without per-lane CHR erases the matrix. Open ?path=multi-cdn per lane.
Fill confused with CRR or Internet DT-out — Confusing CDN miss fill with S3 CRR inter-region replication or with edge-to-client Internet egress invents a fake unit. Opaque CloudFront / S3 talkers need the CE decoder first.
Shield halves both lanes by fiat — Assuming a mid-tier shield automatically halves both lanes without measuring CHR per logo invents free HA. Model shield fill and origin fill as separate Mode C runs — sibling origin-fill estimator. No savings guarantee from adding a second logo.
FinOps checklist
Export client GB and CHR (or miss ratio) per CDN lane.
Estimate fill ≈ client_GB × (1−CHR) per lane; sum carefully for shared origins.
Price edge and origin with dated cites; badge Unknown otherwise. Keep replication (S3 CRR) off the origin-fill line.
Second educational stack (lane CHR sensitivity twin): hold 5 TB on CDN A at 0.80 CHR (≈1 TB fill) while CDN B drops to 0.50 CHR on its 5 TB (≈2.5 TB fill) after a purge — total origin pull ≈ 3.5 TB, not the naïve 2 TB. Re-run Mode C per lane; sibling single-CDN estimator: CDN origin fill. Failover drill remains separate: secondary at ~0.10 CHR for 2 TB clients ≈ 1.8 TB drill fill. Knit CloudFront/S3 origin talkers in the CE decoder; never paste S3 CRR into a fill cell. Synthetic only — rates Unknown; no savings guarantee from a second logo.
Worked-estimate sensitivity (educational)
Hold the 10 TB dual-CDN client baseline (5 TB per lane) and walk three origin-fill runs. Edge $/GB and origin $/GB stay Unknown until dated cites — literacy arithmetic, not a quote or multi-CDN ROI claim.
Run A — Balanced 80% CHR twin: each lane at 0.80 CHR → ≈1 TB fill per lane (≈2 TB total). Open Mode C (?path=multi-cdn) twice; edge-to-client remains a separate Mode A-style line.
Run B — Asymmetric purge twin: hold CDN A at 0.80 CHR (≈1 TB fill) while CDN B drops to 0.50 CHR (≈2.5 TB fill) → ≈ 3.5 TB origin pull, not the naïve 2 TB. Re-run Mode C per lane; sibling single-CDN estimator.
Run C — Failover drill sibling: secondary at ~0.10 CHR for 2 TB drill clients ≈ 1.8 TB drill fill — budget separately from steady-state. Decode CloudFront/S3 talkers with the CE decoder; never paste S3 CRR into a fill cell.
Not necessarily. Two cold caches can increase origin pull versus one warm CDN. Model CHR per lane.
Which calculator mode?
Mode C for origin fill ≈ client_GB × (1−CHR) × origin $/GB. Edge transfer is a separate Mode A-style line.
Is this a vendor bake-off winner page?
No. PipeToll refuses scorecrowns from Incomplete cells. Path literacy only.
Is origin fill the same as S3 CRR?
No. Origin fill is CDN miss traffic toward origin storage/compute. S3 CRR is bucket→bucket inter-region replication. Keep them on separate lines.
How do AWS Cost Explorer lines relate?
When CloudFront / S3 origin talkers appear on an AWS bill, decode them with the CE decoder before Mode C — never paste Internet DT-out into a fill cell.
Does a shielded mid-tier cut origin bill for both CDNs?
Only if both lanes actually miss through the shield. A mid-tier can reduce origin GB while adding another hop to price — model shield fill and origin fill as separate Mode C runs. Sibling: origin-fill estimator.
Can I blend edge $/GB with origin fill?
No. Edge-to-client and origin-fill are different meters (and often different vendors). Keep Mode A-style edge lines apart from Mode C fill — see origin fill.
Trademarks: Amazon Web Services, AWS, Microsoft Azure, Google Cloud Platform, GCP, and Cloudflare are trademarks of their respective owners. PipeToll (Ameti One LLC) is not affiliated with, endorsed by, or sponsored by those companies. Educational comparisons may be outdated — verify current rates with each provider.
