Cloud Storage & CDN Architecture Technical FAQ: GiB vs GB Billing Units, TLS Overhead, and HTTP/3 QUIC

1. Why does my cloud storage invoice report a different gigabyte count than my Linux filesystem (`GiB` vs. `GB`)?

AWS S3, Google Cloud Storage, and Microsoft Azure advertise pricing in "GB-Month," but actually meter storage using **Binary Gibibytes (`1 GiB = 2³⁰ bytes = 1,073,741,824 bytes`)** and **Tebibytes (`1 TiB = 2⁴⁰ bytes = 1,099,511,627,776 bytes`)**. Conversely, network routers and CDN billing meters often use **Decimal Gigabytes (`1 GB = 10⁹ bytes = 1,000,000,000 bytes`)**.

Therefore, $100\text{ TB}$ of decimal files ($10^{14}\text{ bytes}$) equals **$90.95\text{ TiB}$** ($93,132.26\text{ GiB}$) on an S3 storage invoice—a **7.37% binary-to-decimal conversion spread**. Always confirm whether your vendor's rate card defines a "GB" as $10^9$ bytes (Cloudflare R2, Backblaze B2) or $2^{30}$ bytes (AWS S3).

2. If a user aborts a 2 GB download after 50 MB (`HTTP 499`), am I billed for 50 MB or the full 2 GB?

Between the **CDN Edge and the End User**, you are billed strictly for the bytes actually transmitted out of the edge socket into the TCP/QUIC send buffer (~50 MB plus socket window buffer). However, if that request was an **uncached CDN miss**, the CDN edge node may continue downloading the entire 2 GB file from your origin server in the background to populate its cache unless `proxy_Temp_File` or range-slicing is enabled—billing you for the full 2 GB of origin egress! Enable **CDN Byte-Range Slicing** (CloudFront Range Requests / Bunny Slice Caching) on files larger than 20 MB so the CDN fetches the file from origin in discrete 4 MB chunks as the client reads.

3. How much extra bandwidth overhead do TLS 1.3 handshakes and HTTP Response Headers add to my CDN bill?

CDNs meter egress at the **TLS record / HTTP response layer**, which includes HTTP response headers (`~400 to 800 bytes` per response) and TLS 1.3 record framing + certificate chains (`~3.5 KB to 5.2 KB` on new TLS connections, or `~600 bytes` on resumed 0-RTT/session-ticket connections). On a 10 MB video chunk, TLS and header overhead is negligible (`0.05%`). However, on a 2 KB JSON API response over a fresh TLS connection, protocol overhead adds **200% to 250%** to billed byte volume. Enable HTTP/2 and HTTP/3 connection coalescing and `TLS Session Resumption` to amortize handshake bytes across hundreds of requests.

4. Why did my S3 bucket size stay flat in CloudWatch while my monthly storage bill doubled?

In 90% of FinOps audits, this is caused by two invisible storage consumers:

• Noncurrent Object Versions (`S3 Versioning`): When S3 Versioning is enabled, calling `DELETE` only inserts a 0-byte `DeleteMarker` while retaining every overwritten historical copy at full price. Add a `NoncurrentVersionExpiration` lifecycle rule (e.g., expire noncurrent versions after 14 days).
• Orphaned Incomplete Multipart Uploads: Failed chunked uploads do not appear in standard `aws s3 ls` output. Enforce an `AbortIncompleteMultipartUpload` rule set to 7 days.

5. Does S3 Transfer Acceleration (`s3-accelerate.amazonaws.com`) reduce egress costs?

No—**S3 Transfer Acceleration** increases your transfer bill by **$0.04/GB to $0.08/GB on top of standard S3 egress rates**. It is designed strictly to accelerate long-haul *uploads* (`PUT`) from distant clients by routing packets into the nearest CloudFront edge PoP over AWS's private backbone. For downloads (`GET`), use a standard **Amazon CloudFront distribution** instead, which is both faster (via edge caching) and cheaper than S3 Transfer Acceleration.

6. Do `HTTP 304 Not Modified` revalidation responses count toward CDN bandwidth and request charges?

Yes. When a browser sends an `If-None-Match` or `If-Modified-Since` conditional request and the CDN returns `HTTP 304 Not Modified`, no body payload is transmitted, so bandwidth consumption is limited to the `~450-byte` HTTP response header. However, on providers that bill per 10,000 requests (such as CloudFront), an `HTTP 304` is billed at the **exact same per-request rate as a full `HTTP 200` download**. Use longer browser `max-age` values on fingerprinted static assets to avoid unnecessary `304` revalidation traffic.

7. Why am I being billed for `S3 Glacier` restores when I only ran a security scanner or backup check?

Security posture scanners (CSPM tools), DLP scanners, or backup verification scripts that call `GetObject` (instead of `HeadObject`) on objects stored in `S3 Standard-IA` or `Glacier Instant Retrieval` trigger a mandatory **$0.01/GB to $0.03/GB retrieval fee** on every scanned file. Always restrict automated scanners to `HeadObject` metadata inspection or exclude archive prefixes from deep-content scanning.

8. Is data transfer between AWS S3 and AWS Lambda in the same region free?

Yes. Data transfer between an AWS Lambda function and an Amazon S3 bucket located in the **exact same AWS Region** is **$0.00/GB** (provided the Lambda function is either unattached to a VPC or, if attached to a private VPC subnet, routes through an S3 Gateway VPC Endpoint rather than a managed NAT Gateway).

9. How does Cloudflare R2 calculate storage volume—peak usage or daily average?

Both Cloudflare R2 and AWS S3 calculate monthly storage using **Timed Storage (GB-Month / Byte-Hours)**. If you store 100 TB for exactly 10 days of a 30-day month and then delete it, you are billed for $100\text{ TB} \times (10 / 30) = 33.33\text{ TB-Months}$ ($500 on R2). Unlike Wasabi (90-day minimum) or S3 Standard-IA (30-day minimum), R2 Standard and S3 Standard have **zero minimum duration penalties**.

10. What is the fastest way to audit our current cloud storage and CDN TCO across all providers?

Gather four numbers from your last monthly cloud invoice: (1) Active Stored TB, (2) Monthly Outbound Egress TB, (3) CDN Byte Cache-Hit Ratio %, and (4) Monthly `PUT` and `GET` request counts in millions. Enter those four parameters into the Daktan TCO Calculator to generate an instant side-by-side comparison across all 8 major storage and CDN stacks.