It's Not Perfect (yet)
- AKalculator will be perfect (in the distant future)
AKalculator does not brute force through all possible solutions (yet), so it's not always sure that the most optimal solutions are shown. It depends on two high-level algorithms right now, with a greedy approach choosing the cheapest of both solutions.
We have high confidence that regular throughput (<= 1 GiB/s) with regular consumer fanout ratios (<= 3) have close to optimal solutions.
We are actively working on improving the solutions engine and ensuring it finds the best possible outcome.
Open Items
We also track the major open items that result in notable inconsistencies with the computed price. See the Open Items section of the docs for an overview.
The Takeaway?
You should treat this calculator cost more or less as the "maximum" price you'd expect to pay for a cluster, as not only is AKalculator conservative in its estimations but also always doesn't the most cost-efficient solution.
In any case - let this not make you fear the calculator. In my experience, it has never been off more than 15%. The reason for this is that the predominant costs (network, storage) we get right. The question most often relies on which instances do we choose and how many do we choose.
It's very rare that the calculator will give you a cluster that's overloaded with the set parameters and assumptions, because it takes a lot of conservative choices with relation to load.
Subscribe to learn when we make it better!
And don't forget - if you have any feedback/disagreements you are encouraged to discuss it with me and the community in this Featurebase board.