Cloud hosting is not the only way to play. Last year, I moved Procuret from AWS to in-house metal. Here are the results.
By moving off AWS compute-optimised instances onto in-house hardware, Procuret has achieved a 61% mean performance improvement, to a 130ms response time. That’s 61% less time our customers and internal teams spend waiting. The improvement is event starker at the margin - our 99th percentile response time dropped from 500ms to 249ms.
Most dramatically, our API version deployment time dropped from 30 seconds to less than 1 second, enabling us to safely perform releases during the heat of the business day.
Our standby compute headroom, the unused capacity available to respond instantly to a dramatic and unforeseen increase in load, has increased by 2.5x.
In absolute terms, despite the above increases in performance and available compute, our compute costs have roughly dropped in half. And yes, that includes enterprise fibre build out, clean power, racking, machine depreciation, and maintenance.
The truth about cloud compute is that there is no cloud, it is just someone else’s computer in someone else’s building. There is no doubt leasing other people’s computers is a good option in many cases, but as Procuret shows, it doesn’t make sense in every case.
The hardware shown in the header image is for illustrative purposes only, and is not Procuret hardware.