Quorum
Play
How-to guides

Cut the cloud bill

Find the lines that make your Burn, remove idle capacity, pick cheaper instance families, and lower egress without losing Apdex.

Use this guide when Burn is close to income, your runway is short, or you want a better margin before an Exit.

Burn is the Cash you spend per second. It adds up every Node's price, your salaries, egress, and third-party fees. Every cut in this guide has a cost in capacity or risk, so make one change at a time and check Apdex after each one.

Read the bill#

  1. Look at the Cash figure on the HUD. The line under it is your net income per second. A negative value means that Burn is larger than income.

  2. Select the Payments Node. The Inspector opens.

  3. Read the Revenue, Burn, and Salaries rows. Salaries are part of Burn.

  4. Press D to open the Dashboard, once you own Metrics. The Revenue and Burn chart shows when the gap opened. A blue line marks each change you made.

If salaries are most of the Burn, the board is not your problem. Hire fewer engineers, or wait until revenue grows. See Team and velocity.

Find the expensive Nodes#

  1. Select each Node in turn. Every Node card in the Inspector has a Burn row.

  2. Compare the Burn row with the Load row of the same Node.

  3. Mark each Node that costs much and runs under 50% load. These are your first cuts.

A Node with a Standby costs twice its Burn. A Database with Backups costs more than one without. Both are deliberate insurance, so remove them only if you accept the risk.

Remove idle capacity#

  1. For an idle sized Node, open the Instance section and select Smaller (free).

  2. Wait a few seconds of game time. Check that Load stays under 70%.

  3. If you have two copies of a Node and both run cool, sell one. Select it and click the trash button. The refund is half of what the last copy cost.

If the Node goes amber or red after the change, select Undo (Z) at once.

Pick a cheaper family#

  1. Open the Instance section of an App Server, Worker, or Database.

  2. Read the gauges in Resources before you change anything.

  3. Pick a family that matches the full gauge:

    • Compute costs 0.85 times General per vCPU and has half the RAM. Use it when CPU is the limit and RAM has room.

    • Burstable costs 0.4 times General. It runs on CPU credits and drops to a fraction of its CPU when they run out. Use it only for a Node that is idle most of the time.

    • Memory costs 1.3 times General. It is not a cut, but a Database whose working set does not fit in RAM can serve more on a smaller Memory box.

  4. Watch the CPU credits gauge on a Burstable Node. If it falls toward zero, go back to General.

Lower egress#

From the Seed Era, every byte that an App Server sends to the internet costs egress. Media and page views send the most.

  1. Research Compression. Then open the Compression section of each App Server and select Gzip. Page and read responses get smaller for a little more CPU.

  2. Research Content delivery and place a CDN in front of the App Servers. Pages that the CDN answers at the edge cost $0.02 per GB, against $0.09 per GB from an App Server.

  3. Keep Nodes that call each other in the same zone. A call between zones pays egress both ways.

Let capacity follow demand#

Traffic follows a daily cycle. A fleet sized for the evening peak is idle at night.

  1. Research Autoscaling and place an Autoscaler.

  2. Link the App Servers into it.

  3. In Scaling, set a low min and a max that covers the peak. The Autoscaler removes copies when the load falls well under the target.

  4. Later, research Container orchestration. App Servers linked into a Kubernetes Cluster pack onto shared nodes and cost 0.85 times their Burn. Each Cluster also takes Dev Hours until you have a Platform team.

Trim the smaller lines#

  • Analytics. A higher Sample rate earns more Insight and costs more storage Burn. Lower it once you have the Research you need.

  • Auth Provider. It charges per user. At very large user counts, compare its Burn row with the App Server CPU it saves.

  • Search Engine, Cache, Queue. Each size step doubles capacity and Burn. Size them up only when their Load passes 70%.

Check the result#

  1. Look at the Cash line on the HUD. It must move up.

  2. Check that Apdex stays at or above its target. A cut that loses users loses more revenue than it saves.

  3. Open the Dashboard. The Burn line on Revenue and Burn must fall and stay down through the next evening peak.

Practice#

The Out of Credits Sandbox ends with a Burn limit at night traffic. The Read Mostly Sandbox asks you to double capacity under a Burn limit.

NextTeam and velocity