Section 3: Fifty Thousand In, One Hundred Fifty Million Out - Scale Estimation & Workload Analysis

Executive Summary: Derived from viewer behavior, writes come to 50,000 comments per second, an ordinary load. Deliveries come to 150 million per second over 50 million open connections, a 3,000× fan-out driven mostly by large streams. Large streams are also where a per-viewer cap of 10 comments per second keeps that load bounded. Without the cap, the five-million-viewer stream alone would need 25 billion deliveries per second.

The interviewer caps the marker. "Before any boxes. How big is this?" A strong answer sizes two workloads, three orders of magnitude apart, separately. The interviewer's inputs: 50M peak concurrent viewers, 1M concurrent broadcasts, and a largest stream of 5M viewers. That's Aria Vance's v_8812, where Maya and Tomás are watching.

The Write Side: An Ordinary Workload

Most viewers never type, and a few type a lot. Averaged over everyone, assume 1 comment per viewer per 1,000 seconds, about one every 17 minutes.

Peak writes:   50M viewers ÷ 1,000 s      = 50K comments/s
Daily average: ≈ 1/3 of peak              ≈ 16.7K comments/s
Daily volume:  16.7K/s × 86,400 s         ≈ 1.44B comments/day
Mega stream:   5M viewers ÷ 1,000 s       = 5,000 comments/s

Across a million videos, 50K appends a second is routine for a partitioned log. Watch the last line: Maya's "Play the acoustic version!!" is one of 5,000 comments per second on a single video.