Bandwidth budgets: calculate the percentile before choosing a commit

A service priced on a high percentile of measured Mbps cannot be forecast accurately from monthly gigabytes alone. Its charge depends on the distribution of sampled traffic rates and the provider's calculation rule. Obtain the relevant series and contract definition before choosing a bandwidth commitment or explaining an overage.

Leaseweb's metering guidance distinguishes data-volume billing, flat arrangements and burstable percentile billing. For its described percentile method, bandwidth is sampled every five minutes, the highest five percent of readings are discarded and the highest remaining reading determines the rate. Use the method applicable to the actual service; providers can differ in aggregation and charging details.

The purchased port speed, committed bandwidth and billable percentile are separate quantities. A fast port may permit a high short-term rate without making that rate the recurring commitment. Conversely, a modest monthly transfer total does not prove that sustained busy periods will remain below the contracted threshold.

For a hypothetical thirty-day billing period with one valid observation every five minutes, there are 8,640 samples. Five percent is 432 samples. Under the example's explicit rule, sort the samples by traffic rate, discard the 432 highest and use the highest of the remaining 8,208 observations.

Check the actual period rather than assuming every invoice covers thirty days. Confirm the time zone, sampling interval, missing-data treatment and rounding rule. The sample count and discarded population can change with the period or the agreement, so retain the calculation inputs with the billing review.

Also confirm what each observation measures: one direction, an agreed combination of directions, one port or an approved aggregation of services. Summing individual ports' percentiles is not generally the same calculation as summing simultaneous rates first and then taking a percentile. Reproduce the purchased method rather than selecting whichever produces the lower bill.

In illustrative Case A, assume 8,208 readings are forty Mbps, 431 are 180 Mbps and one is 250 Mbps. Discarding the 432 highest removes all the 180-Mbps and 250-Mbps readings. The highest remaining value is forty Mbps, even though the observed peak was 250 Mbps.

In Case B, change one of the forty-Mbps readings to 180 Mbps. There are now 433 readings above forty Mbps. Discarding the highest 432 leaves one 180-Mbps observation, making the example's billable rate 180 Mbps. The change is only one five-minute interval, but it crosses the defined percentile boundary.

This deliberately concentrated example illustrates a discontinuity; it is not a typical traffic pattern or a promise that a fixed number of burst hours is free under every contract. Actual rates may vary continuously. Inspect the observations around the cutoff and test plausible additional busy periods when the measured result is close to a material threshold.

Assume a hypothetical contract has a 100-Mbps commitment costing $600 per month and charges $3 for each Mbps by which the billable percentile exceeds that commitment. Case A costs the $600 commitment because forty Mbps is below it. It does not generate a refund for the unused sixty Mbps under these assumed terms.

Case B exceeds the commitment by eighty Mbps. Its additional charge is eighty times $3, or $240, making the total $840. These rates and the linear overage formula are invented for explanation, not Leaseweb prices or a statement that all providers price bursts in this way.

A different offer might use a tier, another multiplier or a fixed arrangement. Compare the complete quoted alternatives on the same traffic scenarios, including commitment term and any charges outside bandwidth. A higher commit can reduce overage exposure while increasing the amount paid in quiet months.

Identify scheduled events, releases, backups or customer activity that can alter the relevant rate series. Map their expected timing and duration to the service being purchased. A volume forecast can support that analysis, but it cannot substitute for knowing whether the transfer occurs over minutes, hours or several days.

Where an operationally acceptable schedule change is proposed, test its effect on the combined measured series. Moving one transfer may overlap it with another busy workload. Keep application requirements, customer commitments and network performance in the decision instead of treating the lowest possible billing observation as the only objective.

Preserve uncertainty about future workload. Show the cost under a supported base case and under plausible additional busy intervals or growth. Avoid extrapolating a single quiet month's percentile into an annual commitment when the business already knows that larger events will occur later.

Obtain the provider's measured usage and calculate the expected percentile using the applicable rule. Compare it with the invoice's committed rate, measured excess and price. Investigate differences in period, interface scope, missing observations or aggregation before assuming that the quoted rate itself is wrong.

Keep an export of the reviewed period rather than relying only on a live graph that later changes its displayed range. A chart can help locate unusual activity, but the billing calculation should be reproducible from the relevant samples. Document accepted corrections and any unresolved disagreement with the evidence needed for review.

Compare several relevant periods and the business's upcoming workload before changing the purchased commit. Separate a recurring growth trend from a one-off event or a confirmed technical problem. The appropriate commercial response may differ even if both produced the same invoice overage.

Record the chosen commitment, the traffic assumptions supporting it and the conditions that would trigger reconsideration. That gives finance and the operating team a shared explanation of the bill, while keeping rate, volume, port capacity and percentile usage distinct enough to support a real purchase decision.