Dude, Where’s My Throughput? The Downstream Impact of MB vs Mb
We’ve all seen the napkin sketches. A developer calculates that an 8 Mbps stream fits perfectly into a 10 Mbps pipe because, well, 8 is less than 10. But then the buffer wheels start spinning, the AI object detection starts lagging, and the egress bill from the VPC arrives with a comma where a period should be.
The problem isn’t that we don’t know how many bits are in a byte. The problem is that we treat Megabits (Mb) and Megabytes (MB) as interchangeable units in a calculator rather than two different languages. One is for the warehouse (storage), and one is for the highway (delivery).
As we move toward high-stakes environments like real-time emergency response and AI-driven monitoring, the “slop” in our conversions is more than a rounding error. It’s a system failure point.
The Rest vs. Motion Paradox: 1,000 vs. 1,024
Industry veterans likely think in powers of two. But the modern streaming stack is a collision of two different math standards:
- Throughput (Bits in Motion)
Bandwidth is typically calculated in base-10.
A 1,000 Mbps (1 Gbps) stream is exactly 1,000,000,000 bits per second. - Storage & Memory (Bytes at Rest)
Disk and memory use base-2.
A 1 GB file on your server is actually 1,024 Megabytes, or 1,073,741,824 bytes.
When you’re calculating data density for long-term storage, that 7% difference might seem small. But when you’re managing petabytes of VOD content, that’s thousands of dollars in hidden storage costs. What’s more, if you are moving that content across cloud servers, those egress fees could really start to pile up. For a deeper dive into how this impacts your bottom line, check out our previous breakdown: What is a GB? Video Data Density.
The Hidden Costs of Secure Video Delivery
In a secure Virtual Private Cloud (VPC) video doesn’t travel unprotected. It’s wrapped in layers of encryption and transport headers. This adds overhead, and it’s important to consider this Protocol Tax when dealing with secure video delivery.
A 10 Mbps encrypted video stream does not fit inside a 10 Mbps pipe. Transport headers, encryption overhead, control traffic, and retransmissions all consume bandwidth. Without headroom, stability is the first casualty.
Common Protocol Tax & Overhead Table
| Protocol | Tax/Overhead | Ideal Buffer | Impact/Driver |
| SRT | 10% – 25% | 25% | In high-stakes remote monitoring, packet recovery requires additional overhead. |
| RTMP | 5% – 10% | 15% | For standard RTMP ingest, there is a lower tax because it is less resilient than SRT. |
| HLS/DASH (ABR) | 10% – 15% | 20% | Manifest files and segment headers add up over time. More ABR renditions means more VOD files passing through the pipes. |
| Encryption (AES-128) | 1% – 3% | 5% | This is added on top of transport. While negligible per packet, it could become massive at scale. |
Throughput Bottlenecks in AI-Powered Video Workflows
AI models, specifically computer vision models used for real-time object detection, don’t just see a video file. These models consume throughput as they process and analyze each frame.
If an AI engine expects a consistent 15 Mbps to process a 4K feed for campus security, but the network treats it as a bulk file transfer calculated in Megabytes, it may hit a ceiling. Furthermore, AI generates its own Metadata Spine, or a secondary stream of bounding boxes and alerts. These add even more bits to the calculation.
Without accounting for the Metadata Tax, an AI’s real-time insights will suffer from significant drift. This renders them useless in an emergency response or real-time monitoring scenario.
How ABR and Ad Insertion Rely on Throughput
Adaptive Bitrate (ABR) delivery is where the bit-vs-byte battle is won or lost. Players manage their buffers in Bytes (MB), but they receive data in Bits (Mbps).
A common “gotcha” in monetization workflows occurs during Server-Side Ad Insertion (SSAI). Ads are often encoded at different densities than the main feature. If an ad’s bitrate spikes (higher Mbps) and the player’s byte-based buffer can’t fill fast enough, the stream stalls. The ad doesn’t play, the viewer drops off, and revenue disappears. All of this because of a math error between the “bits” on the wire and the “bytes” in the buffer.
Stop Calculating, Start Provisioning
The network doesn’t care about your “perfect world” 8:1 calculations. To build a resilient workflow, you must:
- Account for the Protocol Tax: Always add a 10-20% buffer to your bandwidth provisioning.
- Respect the 1,024 Rule: Use base-2 for storage and base-10 for throughput to avoid egress bill surprises.
- Plan for Metadata: AI and analytics data have a weight that must be budgeted.
Ready to see how your architecture stacks up? Start your Wowza Streaming Engine free trial and test your real-world throughput today.