Techbanq

Cloud · database spend

The database keeps the size it was given on its worst night

Doubling the instance class is the fastest way to end a database incident, and stepping it back down afterwards is a change nobody volunteers to make. Add storage that only accumulates, a licensed engine chosen by default and read traffic that belongs in a cache, and the database tops the bill without one bad decision being taken.

Serving Saudi Arabia from our Riyadh office and our engineering hubs in Lahore and Karachi

Close view of circuit board components under directional light

Information security for cloud delivery

  • ISO

    ISO/IEC 27001

    ISO/IEC 27001 certified

Certifications & registrations

How it got this big

Capacity bought at two in the morning is never handed back

We do not have an office in Saudi Arabia

We serve organisations there from our Riyadh office and our engineering hubs in Lahore and Karachi, travelling for the parts of an engagement that need to happen in person.

The instance class was probably chosen twice. Once at launch, generously, because sizing a database under uncertainty rewards caution. And once during an incident, at two in the morning, when doubling the class was the fastest lever anyone could pull — and it worked, or seemed to, and everyone went to bed. The step back down never happens, because it is a risk with no reward attached: the engineer who proposes it owns the next incident if they are wrong, and owns nothing visible if they are right. So the database keeps its worst night’s size forever, billed hourly.

The incident itself deserves a second look too, because capacity is the mitigation that works on almost everything briefly — including problems that were never capacity. A runaway query, a missing index, a connection storm: each is calmed by a bigger instance and caused by none of its absence. Where the diagnosis never happened, the estate can end up paying permanently for headroom whose only job was to outlast one unexamined bad night.

The drivers

What a managed database charges for besides answering queries

Four drivers, and they are independent: an estate can fix one and keep overpaying on the other three. The review prices each separately before anything moves.

VariableEffect on priceWhy
An instance class held at incident sizePaying every hour for the headroom of the worst nightUtilisation measured over a real window — including the monthly peaks — is the only honest basis for the size. Most databases held at incident size show the gap plainly the first time anyone looks.
Storage that accumulates and never leavesData, indexes and snapshots that only ever growNothing archives old rows, nothing prunes old partitions, and snapshots accumulate on default retention. Managed storage grows in one direction unless lifecycle rules exist — and in most estates, none do.
A licensed engine doing commodity workLicence fees for features the workload never callsCommercial engines earn their fees on the workloads that use what they uniquely offer. Many workloads use nothing the mature open-source engines lack, yet pay the licence line monthly because the engine was chosen by familiarity, not by audit.
Read traffic paying database pricesThe most expensive tier answering the most repetitive questionsWhen the same lookups hit the primary thousands of times an hour, the instance ends up scaled to serve answers that never change between requests. A cache in front of that traffic is almost always cheaper than the capacity bought to absorb it.

Ask any supplier quoting this

  • Which of the engine’s paid features does our workload actually invoke, and where is the evidence?
  • What share of our queries are repeated reads that a cache could answer without touching the database?

The pricing models

Provisioned against serverless: who wins on which traffic shape

Serverless database economics are simple and often mis-sold: bursty wins, steady loses. The shape of your demand decides, not the age of the technology.

OptionA bursty or mostly idle workloadA steady, always busy workloadCost predictabilityThe failure to watch forBest for
Provisioned instancePays around the clock for capacity the bursts use brieflyEfficient — the capacity paid for is the capacity usedA fixed, forecastable monthly numberSized for a peak, then billed for every idle hour after itSteady production loads whose demand is known and measured
Serverless databaseScales towards zero between bursts and charges for the burstsPer-unit pricing keeps the meter running — steady load usually costs more than the instance it replacedA number that moves with demand and can surprise the forecastA traffic spike or a runaway query arrives as a bill instead of an outageDevelopment estates, spiky products and genuine idle time

Our role

What we step down in the database estate and what stays protected

  1. 01

    Measurement before any resize

    Utilisation is measured over a window long enough to contain the real peaks, and any step down is taken one class at a time — observed afterwards, reversible throughout — because the database is the one system where a wrong cut is everyone’s morning.

  2. 02

    Storage that finally leaves

    We put lifecycle rules where none existed — archiving for old data, retention for snapshots, pruning for stale partitions — so the storage line gains, for the first time, a direction other than up.

  3. 03

    Reads moved to their price tier

    The repeated reads go to a cache before anyone scales the instance to serve them, and the licence question gets an audit and a migration estimate — with the decision, and its risk appetite, remaining yours.

How much headroom the business wants to pay for is a comfort level you set; we make sure it is set against measured demand, on a storage line that can shrink, with the cacheable reads already gone.

Common questions

What buyers ask

Can we step the database instance back down without inviting the next incident?+

Yes, if the step down is engineered rather than declared. The preconditions are a utilisation window long enough to contain your real peaks — month-end, campaign days, whatever shape your business has — one class step at a time, an observation period after each, and a rehearsed way back. It is also worth re-opening the incident that bought the current size: if the cause was a runaway query or a missing index, the capacity was treating a symptom, and fixing the actual cause is what makes the smaller size safe. What does invite the next incident is the other approach — cutting by price, on a quiet week’s numbers, with no path back.

Is it worth migrating off a licensed database engine to stop the fees?+

Only the audit can say, and the audit comes first. The question is whether the workload actually invokes what the commercial engine uniquely provides — specific replication modes, procedural dialects, features the application genuinely calls. If it does, the fee is buying something and the migration would be expensive twice. If it does not — and that is a common finding — the arithmetic is a one-off migration cost, which is real engineering with compatibility work, testing and a careful cutover, against a licence line that otherwise recurs monthly forever. Sometimes the honest recommendation after the audit is still to stay; the point is that staying becomes a decision with evidence rather than a default with a fee.

When is a serverless database genuinely the cheaper choice?+

When idle time dominates. Serverless pricing charges per unit of work and scales towards zero between bursts, so it wins wherever a provisioned instance would sit mostly unused: development and test estates that sleep outside working hours, new products whose traffic is genuinely unpredictable, workloads with sharp bursts separated by quiet. On a steady, always-busy production load the same arithmetic reverses — the meter never stops, and the per-unit total usually exceeds the provisioned instance doing identical work. There is also a failure-mode difference worth naming: under a provisioned model a spike degrades service, under serverless it arrives as a bill. Measure the shape of your demand over a real window, and the model chooses itself.

Take the database bill apart with us

A database-spend review that measures real utilisation, prices each of the four drivers on your bill, and names the first step down that is safe to take.

Start the conversation