Competitors
@here Dedicated workers for full flows are now available in alpha for EE. Run each step of a flow on workers dedicated to the flows at blazing speed by just toggling a setting at the flow level and assigning your worker group to that flow. When a worker is assigned to a flow, it is capable of handling every steps. The full flow attached here made of 3 scripts (1 in python, 2 in typescript) and branches executed in 36ms of worker time and took 230ms end-to-end.
This make windmill flows good fit for event streaming cases.
We've benchmarked with @guillaume windmill to be the fastest workflow engine among airflow, prefect, temporal, and by quite a margin for airflow! Read more at: https://www.windmill.dev/docs/misc/benchmarks/competitors
Also next week will be launch week. If you're a keen reader of this channel, you already know almost everything we will present except one major feature on day 5. But we will make special efforts on how we present things by having blog posts explaining in details each feature.
Airflow vs Prefect vs Temporal vs Windmill | Windmill
We compared Airflow, Prefect, Temporal and Windmill with the following usecases:
5 Replies
Its a tricky task to compare your own product to competitors.
I would assume some people will ask of these benchmarks are if you run things in parallel. Said 10 workers and 1000 tasks or something.
Would also at least try to reach out to each platform asking if there are any know configure optimization that should be done. Like on windmill you removed native worker. So you optimization your setup but not competitors, its def a argument that you will get thrown at you
We tried to be as fair as possible. We didn't really optimize our setup more than trying to have the same setup everywhere, 1 worker in each framework. Same ec2 instance so that a basic deployment would use the full resources of the machine. We didn't run anything in parallel. The aim of this is to benchmark the orchestration transition speeds. We tried to be as transparent as possible so that ppl can re-create the benchmarks easily and suggest changes if they find out we were not fair. But at some point, one needs to do benchmark if you want to compare performance.
But if anyone is offering to do independent benchmarking we would love and appreciate it! There will always be accusation of bias if it's coming from us
(but also for the sake of the argument, the results would have been the same with or without native workers, the database had a lot of capacity left and a single idle worker has very very little impact on the db)
Just trying give you a perspective from the outside. I think people will say you optimization it even if the optimization may be so small that you can't measure it. You did something ob your product and not in the others. I understand the change. Outsiders will not.
Other will prop argue that using airflow in sequence is strange and not representative because they use it to process a lot in parallel.. etc.
Just some small feedback before you go live with launch week, then there will hopefully much more people reading it 🙂
It's fair that we should show one benchmark with parralelism to show we both support it and it wouldn't have made a difference, on the contrary
if i'm not mistaken, we did make changes to every setup to be of 1 worker, not just for windmill
@guillaume when we have time, we can try to show parallelism on airflow vs windmill
I like April's opinion when it comes to comparing your product to competition:
https://www.youtube.com/watch?v=-VqmFI9vY7w&t=1153s
Lenny's Podcast
YouTube
A step-by-step guide to crafting a sales pitch that wins | April Du...
April Dunford is a speaker, mentor, podcaster, best-selling author, and beloved returning guest to the show. Last year, she joined me on the pod to discuss product positioning and differentiated value. Today, April offers invaluable insights from her latest book, Sales Pitch: How to Craft a Story to Stand Out and Win. We go deep on the art of ef...