skip to content

Why does reserved parallelism at a browser provider cost the same when your suite sits idle?

level: middleimportance: nice to knowfreq 40%

answer

  1. you bought width, not sessions
  2. capacity, not consumption
  3. no meter starts, none stops
  4. idle width costs what busy width costs
  5. utilisation measures it; a narrower width lowers it

basics

~20 s

Reserved parallelism buys the right to run a set width of sessions at once, so the charge attaches to the reservation itself rather than to any session you open. Idle reserved capacity costs exactly what busy reserved capacity costs.

solid answer

~50 s

A reserved-parallelism arrangement sells you **capacity, not consumption**: the provider holds a fixed width of simultaneous sessions open for your account, and that width is what you pay for whether the veterinary booking suite is running or the practice is closed. The meter never starts and never stops, because there is no meter — the charge is a standing one. That inverts the usual instinct. Shortening a session helps a *metered-time* bill; under a reservation it only returns the width to idleness sooner. The measurement that matches this shape is **utilisation**: how much of the width you hold was ever busy. Packing the nightly regression, the pull-request smoke run and the exploratory reruns into capacity you already pay for lowers your cost per unit of work — but the charge itself moves only when you narrow the reservation, which you cannot size safely until utilisation has shown you the peak.

go deeper

for a junior

Be ready to say what you are actually paying for. Under a reservation you are buying a width of simultaneous sessions, and that charge stands whether the veterinary booking suite runs tonight or not.

for a middle

Be ready to explain why the obvious optimisation fails here. Trimming session duration answers a metered-time charge; under a reservation it only returns the width to idleness sooner, leaving the standing charge untouched.

for a senior

Be ready to argue for scheduling as a cost control. If a reservation is standing, work moved into the hours when the width sits empty adds nothing to the charge — up to the width itself, which still caps how much can run at once — and the reservation's real cost is the width nobody ever used.

for a principal

Be ready to defend the shape itself. A reservation trades a predictable bill for the obligation to keep capacity busy, which is a commitment about how your teams schedule work rather than a purely commercial choice.

## What a reservation actually sells A hosted browser cloud can attach its charge to two quite different things, and the difference decides which optimisation is worth your afternoon. **Metered consumption** attaches the charge to what you used: a session opens, a clock runs, the session is released, and what you owe grows with how long that clock ran. **Reserved parallelism** attaches the charge to what you were promised: the provider undertakes to hold a width of simultaneous sessions available to your account, and that undertaking is the product. Under metering the bill is a consequence of your behaviour. Under reservation the bill is a decision taken in advance, and your behaviour during the period cannot move it. That is why the charge does not fall when the veterinary booking suite is quiet. There is no meter to slow down. The provider is not counting your sessions in order to bill you; it is holding capacity in order to be ready for you, and that undertaking, rather than anything you consumed, is what the charge is for. ## Why idle width is not free The instinct that idle capacity ought to be cheaper comes from metered thinking, where nothing running means nothing counted. Under reservation three things are true at once: - The capacity is committed for the whole period whether you fill it or not, so a quiet week has nothing to hand back. - Your charge is settled before the period begins, so nothing that happens during the period is an input to it. - Width you did not use is not banked for later. The period ends and the unused capacity is simply gone. That last point is the one teams find hardest. Reserved capacity is perishable. A quiet Tuesday does not accumulate into a busier Thursday, and every empty hour is capacity you have already paid for and cannot recover. ## The lever has to match the shape A provider can count elapsed session time, reserved width, or the people granted access, and each of those shapes answers to exactly one lever and slides straight off the others. What matters here is only which lever a reservation answers to, and duration is emphatically not it. Applied to the veterinary booking suite: trimming the appointment-calendar journey so it finishes sooner is a genuine improvement, and under a metered-time charge it is also a saving. Under a reservation it is not a saving at all. It returns the width to idleness sooner, and idleness is precisely what you are already paying for. ## Utilisation is the measurement; the width is the lever Under reservation the useful question changes. It stops being *how long did that session take* and becomes *how much of the width we hold was ever busy*. That reframing produces a very different set of actions: 1. **Pack the width.** Move work you have been deferring into the reservation — the exploratory reruns after an unstable night, the long vaccination-reminder journey nobody wants blocking a pull request, the browser sweep the team keeps postponing. Under a standing charge, work moved into unused width costs nothing extra. The width is still an admission ceiling, though: what you fill is the space inside it. 2. **Schedule against the trough.** If the nightly regression run leaves the width empty all morning, the morning is the cheapest capacity you own. Pull-request runs, ad-hoc reruns and demo preparation belong there. 3. **Right-size deliberately, and only then.** Once you know how much of the width is genuinely busy at the peak, narrowing the reservation is the only change that moves the charge — and it should be made from evidence rather than from one quiet month. The order matters. Narrowing the reservation is the only real saving available, but narrowing it before you know your peak trades a predictable bill for a wait at the worst possible moment. ## What reservation is really buying Reservation is not a worse deal, just a different one, and what it sells is **predictability**. The charge is known before the period starts. It does not move when the veterinary booking suite grows a new regression pack, when an unstable week fills the schedule with reruns, or when someone leaves a long journey running over lunch. A metered charge makes the opposite trade: you pay for what you used, which is fairer in a quiet month and alarming in a bad one. The two shapes therefore reward opposite disciplines: - Metered time rewards **frugality** — never open a session you do not need, and release the ones you have as soon as the work is finished. - Reserved parallelism rewards **density** — never let the width you already bought sit empty, because empty width is waste you have already paid for. A team that carries metered habits into a reservation will optimise energetically, watch nothing happen, and conclude that browser-cloud spend is simply immovable. The actual failure was upstream: they chose a lever without first establishing which quantity the bill is a function of.

  • If a reservation is standing, what work should you deliberately move onto it?
    Anything that currently waits for a quieter moment: the exploratory reruns after an unstable night, the long vaccination-reminder journey nobody wants blocking a pull request, the browser sweep the team keeps deferring. Under a standing charge, work moved into unused width costs nothing extra, so anything that fits inside the width you already hold is free at the margin. The width itself is still a hard ceiling: sessions asked for above it wait or are refused by the far side, which is why the aim is to fill the reservation rather than to treat it as unlimited.
  • What does a reservation make predictable that metered time does not?
    The bill. A reservation is known before the period starts and does not move when the veterinary booking suite grows a new regression pack or when an unstable week fills the schedule with reruns. Metered time makes the opposite trade: you pay only for what you used, but a runaway suite or a stuck teardown shows up as a charge nobody forecast. Predictability is the thing a reservation is actually selling.

A reserved width is a rented parking bay, not a taxi fare. The bay costs the same overnight as it does at midday, so parking more cars in it lowers the cost per car — but the only way to lower the rent is to rent a smaller bay.

saying these in an interview costs you the question

  • Says shortening every session lowers a reserved-parallelism charge.
  • Assumes any browser-cloud charge starts when a session opens.
  • Treats unused reserved width as free because nothing ran.
  • Believes a quieter week automatically produces a smaller reservation charge.
  • Thinks utilisation and duration answer the same shape of charge.