A sample of what the cost review hands you.
Sample Utility Co. does not exist. Its usage, its changes and its savings are invented to show the form of the report. The method and the techniques are real.
The cost map
Eight kinds of usage, and the share each one took.
The map comes first. Our baseline was within 0.6% of the client's own figure.
| Kind of usage | Usage in one month | Share of the cost |
|---|---|---|
| Batch pipelines | 412.0 million compute-seconds | 58% |
| Ontology indexing | 96.0 million compute-seconds | 13% |
| Storage | 1,450,000 gigabyte-months | 8% |
| Interactive analysis | 54.0 million compute-seconds | 8% |
| Ontology queries and actions | 38.0 million compute-seconds | 5% |
| Ontology volume | 52,000 gigabyte-months | 4% |
| Live deployments | 14.0 million compute-seconds | 2% |
| Code Workspaces | 9.0 million compute-seconds | 1% |
A compute-second is the platform's unit of work. A gigabyte-month is one gigabyte held for one month.
Change 1 of 11, the largest
Three years of history, read every hour. One change, in full.
- What we saw
The transform read three years of outage history on every hourly run, to add one hour of new events.
- What we changed
It now reads only the new events. A full rebuild on a branch gave an equal output.
- The measured result
Usage fell from 88.0 million to 40.0 million compute-seconds each month, a fall of 55%.
- The guardrails
Four, named before the change: how old the outage data is when the notification service reads it, how long a notice takes on a replay of the storm day, how many rows differ from the unchanged output, and the share of notices delivered inside 10 minutes. All four held.
- The approval
The same outputs in a cheaper way, so the platform lead approves. The second engineer reviewed the change, and verified the saving.
Four more changes
Four more changes, each measured.
One engineer made each change. The other verified the saving.
| What we saw | What we changed | Usage fell by |
|---|---|---|
| A pipeline rebuilt 84 times a week for a morning reader | It runs once, at 05:00 on working days. The head of vegetation management accepted the new schedule in writing | 72% |
| A job with 64 executors, 14 in use | Stepped down to 16, one level at a time, and proved on the busiest day | 50% |
| One column made every customer change each night | The column was removed. The nightly change fell from 5.2 million rows to about 9,000 | 61% |
| A dashboard joined 400 million rows on every open | The join runs once each morning, and the dashboard reads the small result | 74% |
The guardrails
Twelve guardrails, measured before and after. Five are shown here.
A guardrail is a service level or a business measure that must not get worse. A saving counts only when every one held.
| Guardrail | Before | Limit | After |
|---|---|---|---|
| The age of outage data when the notification service reads it | 14 minutes | 15 minutes | 12 minutes |
| An outage notice, from the source event to the delivered message, on a replay of the storm day | 9.2 minutes | 10 minutes | 8.6 minutes |
| Inspection images available after upload | 2.6 hours | 4 hours | 3.1 hours |
| The load time of the vegetation planner's dashboard | 38 seconds | 38 seconds | 4 seconds |
| Outage notices delivered inside 10 minutes | 97.2 percent | 97.2 percent | 97.9 percent |
One guardrail moved toward its limit and stayed inside it. The report shows that too.
Where the claim stops
Four things the statement does not count.
- A change that failed its proof
On a replay of the busiest storm day, a smaller compute setting took the notification time to 11.4 minutes, against a limit of 10. We rolled it back on the same day. Its saving is not counted.
- A change the owner declined
The head of grid operations declined a slower outage feed. It would break the notification target.
- A forecast
Six more opportunities are open, each sized as a range. None is part of the verified saving. Two wait for a decision by a business owner.
- The future
The statement records usage that was measured. It is not a guarantee of future cost: a later change in volume, in use or in price changes the cost.
How the gain is held
Four controls stay behind, and one more measurement.
- Budgets and alerts
Usage accounts by owner, each with a budget and an alert.
- Large compute settings
Restricted, with a named approver.
- A weekly sweep
Each recommendation is assigned to an owner.
- The change procedure
Three cost questions are added to it.
- The second measurement
Quant8 measures the changed resources again later, and reports any saving that was lost, with the cause.
Ask what yours would show.
Tell us how you pay for Foundry and who sees the figure. We start with the cost map.
Book a briefing James replies to you himself.