Summary

  • Datadog's high watermark plan discards the highest 1% of hourly host-count readings and uses the maximum of the remaining lower 99% as the monthly billable host count.
  • In our invented 720-hour month, twenty readings at 400 hosts and seven hundred at 100 leave that count at 400. A hypothetical hybrid contract with a 100-host monthly minimum produces 6,000 excess host-hours instead, alongside the minimum. Those are quantities to price, in different units.

An engineering team can retire its temporary servers on schedule and still leave procurement with a large monthly number to explain. The machines have finished their work. The billing formula has finished a different job: turning their visit into the quantity the customer bought.

Datadog documents two plans for host-based products. The high watermark plan, HWMP, extracts a monthly host count from hourly readings. The hybrid monthly/hourly plan, MHP, combines a monthly minimum with hourly charges for host-hours above that commitment. The choice buys a way of remembering the workload, as well as a price.

That matters to the people who made the fleet elastic in the first place. Being able to remove capacity promptly is an operational achievement. How much of its short life survives in the monitoring quantity depends on the contract.

Twenty hours with a long commercial shadow

For Infrastructure Monitoring, Datadog defines a host as a physical or virtual operating-system instance. It records the number of unique monitored hosts each hour. Under HWMP, the highest 1% of those hourly readings is excluded, and the maximum of the lower 99% becomes the monthly billable count.

The exclusion softens the very highest end of the month. A temporary expansion can last beyond that excluded portion, leaving enough high readings to set the retained peak.

Build a deliberately invented example: a thirty-day month with 720 hourly readings. Seven hundred readings show 100 monitored hosts; twenty show 400. No customer's usage or invoice is being presented here.

Remove seven or eight of the highest readings and there are still readings at 400. The HWMP metered monthly count is therefore 400. The result survives both rounding choices, avoiding a dependency on edge rules the public wording does not spell out.

The extra 300 hosts in this example lasted twenty monitored hours. The contract turns that history into a monthly quantity. A low monthly average, prompt retirement and a high retained billing count can all describe the same fleet.

The other unit keeps the duration

Datadog's MHP charges a minimum monthly commitment, then an hourly rate for host-hours above it. Give the same invented workload a hypothetical minimum of 100 hosts. The excess is 300 hosts for twenty hours: 6,000 host-hours. The monthly 100-host commitment remains.

Plan applied to our invented workload Quantity to price
HWMP 400 monthly billable hosts
MHP with an assumed minimum of 100 A 100-host monthly minimum plus 6,000 excess host-hours

Putting 400 beside 6,000 makes a wonderfully quick comparison of incompatible units. One counts hosts; the other accumulates time, with a separate minimum still attached. The missing inputs are the buyer's actual order prices, commitments and monitored-host trace, which could change which structure is economical.

A stable estate and an estate that grows briefly present different shapes to the supplier. The hybrid formula makes duration above the minimum part of the excess quantity. The high watermark formula makes the retained high count the monthly measure. Neither choice happens automatically because an engineer described the fleet as elastic.

Read the CSV's unit before adding its rows

Datadog's Bill Overview documentation offers a useful check. For organizations billed at the 99th percentile, the resource CSV gives each host a Usage Value of 1 and identifies the relevant measurement hour in Timestamp. For organizations billed on a sum basis, the same value column contains host-hours. Datadog's example for a host present throughout a thirty-day month is 720.

The heading stays familiar while the unit changes. A tidy inventory on the final day cannot explain an earlier peak-setting hour by itself. A monthly average may be accurate and still answer a different question from the invoice's host quantity.

Product boundaries matter too. Datadog documents sum, average, maximum and HWMP aggregation, together with included, committed and allotted usage. Logs, containers and custom metrics have their own rules. Our calculation concerns the documented host plans. The published default uses calendar-month usage unless the order states otherwise.

Elastic infrastructure gives a team the ability to send machines home when the work ends. Procurement chooses what record of that work remains chargeable. Bringing the hourly curve into the buying decision gives the budget a chance to enjoy the same flexibility as the fleet.

Sources