Cockroach Labs Estate economics What idle nodes, provisioned bytes and peak-sized bills cost, worked out from your estate.
  1. Nodes 15 baseline
  2. Avg CPU utilization 15.0% idle hardware
  3. Storage provisioned 300 GB 50 GB of data
  4. Est. monthly saving 0%

Step 8 of 8 - Summary

Everything you changed, added up

Seven pages of controls come down to one number: what this estate would pay against what it pays today. Here's where that number comes from, and what you haven't switched on yet.

Virtualization One Private Host Cluster
Storage Plenum Tier 80% of data as cold

Estimated saving against your estate today

nothing switched on

Every switch above is off, so this estate is priced exactly as you described it on the front page. Throwing all three would take 58.7% off that bill, 53.2% from virtualization and 5.5% from storage.

0%25%50%75%100%Your estate todayseparate clusters, block storage100.0%Your estate tomorrownothing switched on yet100.0%Your estate today against your estate tomorrowAgainst a baseline of separate clusters on block storage, this estate is set up to pay 100.0% of that bill: a saving of 0.0%, of which 0.0 points come from virtualization and 0.0 points from storage. Throwing the switches still off would take off a further 58.7%, for 41.3% of the baseline in total.
From virtualization
0.0%
From storage
0.0%
Still on the table
58.7%
How this is added up
  • The baseline is the estate you described on the front page, bought the way it's bought today: 5 separate clusters on block storage. Every percentage on this page and every percentage in the bar at the top of the screen is measured against that one number.
  • These switches aren't independent - the storage layer changes what a node costs, and consolidation changes how many nodes there are to charge for - so a total can only be split along some order. This one starts at the baseline and climbs through the switches you've already thrown before it reaches the ones you haven't. That's what makes the switched-on rows add up to exactly the 0.0% in the bar above. Each remaining row is then priced from the row above it rather than from the baseline, so they add up to the 58.7% that's left rather than to more than there is: throw one on its own and it's worth more than the line here says, because it would be collecting overlap the row above it has already claimed.
  • Nothing on this page prices the rest of the argument. Fewer bad days from a runaway neighbour, a scale-out that finishes in seconds instead of hours, and a node loss that doesn't move data are all worth something, and none of them are worth a percentage this model can defend.
  • The rates behind the arithmetic are derived from a design-doc cost model at 200-node scale, not from list prices and not from final pricing. Illustrative, not measured. The claim is the ratio, whatever the bill turns out to be.

Every figure here follows the estate you described. Change it and this page changes with it - or go back through the pages and read why any one of these switches is safe to throw.