Tail Latency

Definition:

Tail Latency refers to the slowest response times in a system - the “long tail” of the latency distribution. While most requests are fast, a small percentage (like the slowest 1% or 5%) can be dramatically slower, creating a poor user experience even when average performance looks good.

Why It’s Called “Tail” Latency In a graph of response times, most requests cluster on the left (fast), but some stretch out to the right forming a “tail” of slow requests.

Number of Requests
     ▲
     │    📊
     │   ████
     │  ██████
     │ ████████
     │██████████ 🐍 ──── Tail (slow requests)
     └─────────────────────────► Response Time
     Fast ←            → Slow

Example:

• Amazon’s Discovery

Tail latency increase of +100ms = -1% in sales
Why? That 1% of slow users represents:
├─ 1 million daily visitors
├─ 10,000 affected customers
└─ $50,000+ daily revenue loss