Re:Cloudflare Workers Now Even More Unbound: 15 Minutes, 100...
by vlovich123 on Friday November 19, 2021 @05:54PM
(#29283670)
The included amounts in your monthly fee need to be priced in to get an appropriate estimate. I would compute your monthly cost for a certain number of requests rather than trying to multiple a per request number because none of the slopes intercept at 0. At 1 million requests per month, you'll likely be paying $5 on either plan since that's below the baked-in thresholds.
I wouldn't really compare the pricing because they're different use cases.
* Bundled workers can't do 100ms of compute per request. They're limited to a max of 50ms.
* Unbound scripts can use as much or as little compute as you want per request up to a maximum wall time of 30s.
* Cron triggers can use up to 15 minutes of wall time per invocation (but can't handle requests).
If you can do all I/O & CPU within 50ms, then unbundled might theoretically be cheaper still but that's just on my coarse napkin math. It's hard to correctly model how many Workers you'll end up needing during the month which provides large error bars (e.g if you get 100 requests per second across the globe, the cost will be very different than if you got 100 rps from just 1 machine for the month).
The option to pick is the one that satisfies the shape of the problem you are using. Predicting the price you'll end up paying is a bit hard to estimate correctly vs just seeing what your bill ends up being.
I wouldn't really compare the pricing because they're different use cases.
* Bundled workers can't do 100ms of compute per request. They're limited to a max of 50ms.
* Unbound scripts can use as much or as little compute as you want per request up to a maximum wall time of 30s.
* Cron triggers can use up to 15 minutes of wall time per invocation (but can't handle requests).
If you can do all I/O & CPU within 50ms, then unbundled might theoretically be cheaper still but that's just on my coarse napkin math. It's hard to correctly model how many Workers you'll end up needing during the month which provides large error bars (e.g if you get 100 requests per second across the globe, the cost will be very different than if you got 100 rps from just 1 machine for the month).
The option to pick is the one that satisfies the shape of the problem you are using. Predicting the price you'll end up paying is a bit hard to estimate correctly vs just seeing what your bill ends up being.