Enterprise Bandwidth Procurement: 95th Percentile Burstable Billing vs Per-Terabyte Commit Contracts

How 95th Percentile (5-Minute Sampled) Billing Works

In a standard 30-day billing month ($43,200\text{ minutes}$), carrier routers and CDN edge aggregators poll outbound byte counters every **5 minutes (300 seconds)**, generating exactly **8,640 discrete bandwidth samples**:

The 36-Hour Threshold Trap: Under 95th percentile billing, if a DDoS attack or viral software release spikes your bandwidth from a baseline of `2 Gbps` up to `40 Gbps` for **35 hours** during the month, that entire spike falls inside the discarded top 5% window and costs **$0.00 extra**. However, if the spike lasts **37 hours** (just 1 hour past the 36-hour threshold), Sample #433 jumps to `40 Gbps`, and you are billed at the `40 Gbps` rate for the *entire month*!

Converting `$/Mbps` to Effective `$/GB`: The Utilization Load Factor

At 100% flat continuous utilization over a 30-day month ($2,592,000\text{ seconds}$), $1\text{ Mbps}$ of bandwidth transfers exactly:

Rate Conversion Matrix: `$/Mbps` vs. Effective `$/GB` by Workload Profile

Real-world consumer traffic follows a diurnal sine wave (peaking in the evening and dropping overnight), resulting in an **Average-to-95th Utilization Factor ($U$)** between `0.25` (spiky events) and `0.55` (global SaaS):

Auditing Carrier 95th-Percentile Invoices with PromQL & SQL

To verify your carrier or CDN provider's monthly 95th-percentile invoice against your own Prometheus SNMP exporter (`ifHCOutOctets`) or edge log database, compute the 5-minute rate across a 30-day (`8,640`-bucket) window:

# 1. Prometheus PromQL: Exact 30-Day 95th Percentile in Mbps (5-Minute Subquery)
quantile_over_time(
  0.95,
  (rate(ifHCOutOctets{ifName="et-0/0/0"}[5m]) * 8 / 1e6)[30d:5m]
)

# 2. PostgreSQL / TimescaleDB: Sample #433 Audit from 5-Minute Byte Buckets
WITH five_min_buckets AS (
  SELECT
    time_bucket('5 minutes', timestamp) AS bucket,
    (SUM(bytes_sent) * 8.0) / (300.0 * 1000000.0) AS mbps
  FROM cdn_edge_logs
  WHERE timestamp >= '2026-09-01 00:00:00Z' AND timestamp < '2026-10-01 00:00:00Z'
  GROUP BY 1
)
SELECT
  percentile_disc(0.95) WITHIN GROUP (ORDER BY mbps ASC) AS billable_95th_mbps,
  AVG(mbps) AS average_mbps,
  ROUND((AVG(mbps) / NULLIF(percentile_disc(0.95) WITHIN GROUP (ORDER BY mbps ASC), 0))::numeric, 3) AS utilization_factor
FROM five_min_buckets;

Multi-CDN Peak Shaving Architecture

When operating a primary contract billed on 95th percentile `$/Mbps` alongside a secondary CDN billed on flat `$/GB` (or zero-egress R2/Cloudflare), automated traffic controllers monitor real-time throughput and execute **Peak Shaving**:

• Baseline Fill (`0 to Commit_Mbps`): Route 100% of traffic through the 95th-percentile pipe up to your committed `Gbps` floor so your Average-to-95th Utilization Factor approaches `0.90+`.
• Overflow Spillover (`> Commit_Mbps`): Whenever real-time 5-minute SNMP telemetry exceeds your committed `Mbps` ceiling, dynamically shift excess DNS weight to your flat-rate per-GB CDN (e.g., Bunny Volume Tier or Cloudflare R2). This keeps your 95th percentile meter flat while paying only per-gigabyte rates on temporary traffic spikes.