Bandwidth Commit vs Burst on Dedicated Servers: 95th Percentile Billing Explained

Commit vs Burst: The Line That Decides Your Invoice

When a dedicated server plan says “100 Mbps unmetered” or “1 Gbps with 10 TB included,” two different billing models are hiding behind the marketing. Committed bandwidth is a guaranteed rate you pay for 24/7, billed at the 95th percentile or flat. Burst bandwidth is short-term capacity above your commit that the provider lets you consume, usually billed per gigabyte or throttled once you exceed it. Choosing between them is a forecasting problem, not a hardware problem, and getting it wrong either wastes 40% of your budget on idle capacity or blows it up during a launch event. This guide gives you the math for both models.

If you are still comparing providers on total cost, the dedicated server comparison lists which plans bill per-commit and which bill per-and-overage — that distinction changes a 12-month cost model by several thousand dollars.

How 95th Percentile Billing Works

Most tier-1 transit and colocation contracts bill on the 95th percentile (95th pct), measured in five-minute samples at the switch port. In a 30-day month there are 8,640 five-minute intervals. Drop the top 5% — the highest 432 intervals — and the next-highest sample is your billable rate. The practical consequence is important for game hosts:

• A single 15-minute traffic spike is fully discarded. 95th pct forgives three consecutive 5-minute bursts, which is exactly why it suits spiky game traffic.
• A sustained 90-minute spike of any magnitude will show up, because you have exhausted the free samples.
• Your monthly invoice is therefore not the sum of bytes transferred — it is the level you cannot exceed for 5% of the month.

This is why two servers moving the same 30 TB per month can pay wildly different amounts: one spreads traffic evenly at 90 Mbps, the other peaks at 400 Mbps every evening and idles at 20 Mbps overnight. Both use the same data, but the second one’s 95th pct is far higher.

Traffic pattern	Monthly data	Peak 5-min	95th pct rate	Bill @ $0.60/Mbps commit
Flat distribution	30 TB	95 Mbps	90 Mbps	$54.00
Evening-peaked	30 TB	400 Mbps	320 Mbps	$192.00
Game launch spike (2 h)	30 TB	900 Mbps	140 Mbps	$84.00

Note the third row: a two-hour launch spike on an otherwise quiet server barely moves the 95th percentile because 2 hours is under the 5% discard window of a 30-day month. A four-hour spike would not be forgiven.

How Burst and Overage Billing Works

The alternative model is a base rate plus burst allowance. Typical structures you will encounter:

• Included transfer, then overage. “10 TB included, $5 per additional TB.” Simple to forecast if your traffic is predictable; brutal if a DDoS or a viral moment pushes you 50 TB over in one day.
• Burst-above-commit at a premium rate. You commit to 100 Mbps, and any 5-minute sample above 100 Mbps is billed at, say, $3/Mbps of burst. Honest but requires active monitoring to avoid a surprise.
• Rate-limited burst. You commit to 100 Mbps and the port allows up to 1 Gbps, but traffic above commit is dropped or shaped once a threshold is hit. No surprise bill, but player experience degrades at exactly the wrong moment.

For a dedicated game server, the rate-limited model is the most dangerous. A dropped packet during a competitive match is worse for retention than a higher hosting bill. If you plan to burst regularly, negotiate burst-above-commit with a defined cap rather than accepting silent shaping.

Sizing Your Commit: The Two-Week Method

Do not size a commit from an average. Sample your actual interface counters every 5 minutes for at least two weeks, including one peak weekend, then compute:

1. Baseline commit = your 95th pct rate over that period, rounded up to the next 10 Mbps.
2. Burst headroom = your maximum 5-minute rate times 1.3, to absorb the launch or event you have not had yet.
3. Overcommit factor = max rate ÷ 95th pct rate. If this is above 4×, you are a burst customer and should pay for burst, not for a high flat commit.

For a game server, run the counters yourself rather than trusting the provider’s graph. On Linux, collect them every 5 minutes:

# /etc/cron.d/bw-sample — awk extracts Mbit/s over the 300s interval
*/5 * * * * root /usr/bin/awk 'NR==1{r=$2} END{printf "%.1f\n", (r+0)/(300*125000)}' /sys/class/net/eth0/statistics/tx_bytes > /var/log/bw/$(date +\%s).mbit

Sort that log at the end of the month, drop the top 432 entries, and you have your exact 95th percentile — the same number your provider will bill on. Compare it against their invoice. Discrepancies in the 95th pct calculation (some providers sample per-second, some per-minute; some measure in+out, some out only) are a common source of billing disputes, and having your own data settles them.

Game-Specific Traffic Profile

Per-player bandwidth establishes your floor. Approximate sustained rates per connected player, including overhead:

Game / mode	Per-player in	Per-player out	Notes
Minecraft (vanilla, 20 tick)	15-30 Kbps	40-80 Kbps	Outbound dominates; chunk generation on join causes big spikes
Minecraft (modded, 200+ mods)	40-90 Kbps	120-250 Kbps	Asset sync on first join can hit 5-10 Mbps per player briefly
CS2 competitive	30-45 Kbps	60-90 Kbps	Both directions rise with tick rate
ARK: Survival Ascended	50-120 Kbps	100-200 Kbps	Highest of common survival titles; mods multiply it
FiveM / GTA RP (32 slots)	80-150 Kbps	250-500 Kbps	Custom asset streaming dominates outbound

Multiply the outbound figure by player count and add 20% for retransmits and DDoS-scrubbing overhead. A 200-player modded Minecraft server at 250 Kbps outbound per player needs roughly 60 Mbps sustained — which sets your commit floor. A 500-player FiveM server at 400 Kbps needs 240 Mbps, and that is before asset downloads spike the port.

The Decision Rule

• Predictable, steady traffic (a long-running community with stable player counts) → buy a commit sized to your 95th percentile. Cheapest per Mbps, and 95th pct washes out short spikes.
• Event-driven traffic (season launches, tournament weekends, streamer raids) → low commit plus negotiated burst-above-commit with a hard cap. Never accept silent rate-limiting.
• Asset-heavy servers (FiveM, modded Minecraft, custom map servers) → model the CDN separately. Serving 5 GB of custom assets to 100 joining players is 500 GB of transfer in an hour and will dominate any commit model. See the storage tiering guide for game asset CDNs for that side of the equation.
• DDoS-prone titles → confirm whether scrubbed traffic counts toward your billable rate. On some plans it does, which means an attack costs you money even though players never notice it.

The most common failure is buying an unmetered plan when you actually need burst capacity, or a metered plan when your traffic is flat. Both cost real money every month. Sample first, then commit.

If you would rather avoid bandwidth forecasting entirely for a single high-traffic game server, a dedicated plan with a generous included transfer allowance and a price that never rises removes the modeling risk. InterServer’s dedicated servers are price-locked for the life of the plan, and new customers can use promo code TRYINTERSERVER for a $0.01 first month to measure real 95th-percentile traffic on the actual port before committing.