You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We need more tests and benchmarks. They will help us debug and improve the code quality.
This issue is for newcomers of this project. If you would like to contribute, but don't know which work you can do, you can start with a test or a benchmark. The experience with compio will help you get familiar with the code.
I have a question, I am currently trying to decided between monoio and compio. The challange I face with monoio is largely related to some hard cross compilation issue.
I sincerely appreciate these initial benchmarks. However, are there any benches related to scaling per additional core?
For example, monoio has some nice benchmarks on its Readme with 1, 4, and 16 cores that clearly showcase the near linear scalability while preserving low latency.
I am asking because I am experimenting with some kind of MPP architecture for message streaming and in that case, latency, and scaling matters quite a bit.
The challange I face with monoio is largely related to some hard cross compilation issue.
AFAIK monoio doesn't support Windows. Cross-platform is a key feature of compio.
However, are there any benches related to scaling per additional core?
Not now. I have noticed the nice benchmarks in monoio, but I didn't have time to write a similar one. That's what this issue is for: calling newcomers to contribute compio starting from adding more tests and benchmarks. PRs are welcome.
You don't need to implement the client yourself (L31 -- L41). The repo has already got a testing client.
Use read_exact and write_all just like the monoio server.
The dispatcher is a suitable way here to write a thread-per-core server. However, I'm not sure which one should be better, either a single-threaded server, or a thread-per-core server. You can write both with compio and benchmark them.
Hello, I’d like to adapt a Hyper-based HTTP Client library for use with compio. This will involve porting much of Hyper’s underlying code, and making special adjustments for compio where necessary. Do you think it can work similarly to how it does with Tokio? I’m new to compio and not very familiar with its development progress yet.
No, they are for different purposes. Those on docs.rs are API docs. The website is a book describing the ideas and (hopefully) providing a quick start.
I added compio feature to my web framework and postgres driver to see the difference between it and tokio-uring as runtime.
The benchmark has been submitted to techempower and you can check the recent score here.
If I'm reading correctly, it looks like compio == tokio < tokio-uring(by ~5%)` ?
Sorry I should have add more context to it as the bench site does not offer individual test based filter.
The xitca-web [barebone] can be ignored as it's tokio only and not realistic.
xitca-web runs on tokio-uring and xitca-web [compio] runs on compio. Both are almost identical implements on the http side.
The postgres drivers are different. For tokio-uring I took advantage of it's compat to tokio ecosystem and the driver is written in the way that can run on tokio entirely. For compio the driver is rewritten for io_uring specifically. Therefore I consider them both are as efficient in their own way.
From my observation tokio-uring and compio are at the same level of perf for the specific bench workload techempowers offers. That said I may not have either perform in their peak so there could be rooms left for either for improvement.
i have run compio in prod for sometime for large service. from my experience compio yields same results as tokio, no real advantages. i dont run it anymore, it may perform better in latest versions though
Activity
fafhrd91 commentedon Oct 30, 2024
I added compio as part of ntex benchmarks to tech empower
test run example is here
marvin-hansen commentedon Feb 24, 2025
I have a question, I am currently trying to decided between monoio and compio. The challange I face with monoio is largely related to some hard cross compilation issue.
I sincerely appreciate these initial benchmarks. However, are there any benches related to scaling per additional core?
For example, monoio has some nice benchmarks on its Readme with 1, 4, and 16 cores that clearly showcase the near linear scalability while preserving low latency.
I am asking because I am experimenting with some kind of MPP architecture for message streaming and in that case, latency, and scaling matters quite a bit.
Berrysoft commentedon Feb 24, 2025
AFAIK monoio doesn't support Windows. Cross-platform is a key feature of compio.
Not now. I have noticed the nice benchmarks in monoio, but I didn't have time to write a similar one. That's what this issue is for: calling newcomers to contribute compio starting from adding more tests and benchmarks. PRs are welcome.
marvin-hansen commentedon Feb 24, 2025
@Berrysoft Thank you,
I have drafted an initial benchmark based on the monoio bench to compare Tokio, monoio, and compio.
it is a simple ping-pong task where the client calculates how many responses it got from the server.
However, as I am new to compio I don't know if that is the best way to implement the task.
Would you please take a look and give some feedback to make the implementation better?
https://github.com/marvin-hansen/monoio-benchmark/blob/master/compio-server/src/main.rs
Berrysoft commentedon Feb 25, 2025
Two comments:
read_exactandwrite_alljust like the monoio server.The dispatcher is a suitable way here to write a thread-per-core server. However, I'm not sure which one should be better, either a single-threaded server, or a thread-per-core server. You can write both with compio and benchmark them.
0x676e67 commentedon May 27, 2025
Hello, I’d like to adapt a Hyper-based HTTP Client library for use with compio. This will involve porting much of Hyper’s underlying code, and making special adjustments for compio where necessary. Do you think it can work similarly to how it does with Tokio? I’m new to compio and not very familiar with its development progress yet.
George-Miao commentedon May 27, 2025
@0x676e67 You can checkout compio-rs/cyper and see if that fits your need
0x676e67 commentedon May 27, 2025
Nice, the compio-rs/cyper project showed me that this is possible.
adamanteye commentedon Dec 8, 2025
Hi, I didn't find the git repository behind compio buffers. Are these documentations generated from the rust code?
Berrysoft commentedon Dec 8, 2025
The website source is in another repo: https://github.com/compio-rs/compio-website
adamanteye commentedon Dec 8, 2025
Will compio-website serves as the doc site of compio instead of its crates on docs.rs for the users?
Berrysoft commentedon Dec 8, 2025
No, they are for different purposes. Those on docs.rs are API docs. The website is a book describing the ideas and (hopefully) providing a quick start.
fakeshadow commentedon Jan 2, 2026
I added
compiofeature to my web framework and postgres driver to see the difference between it andtokio-uringas runtime.The benchmark has been submitted to techempower and you can check the recent score here.
kkimdev commentedon Jan 2, 2026
If I'm reading correctly, it looks like
compio==tokio<tokio-uring(by ~5%)` ?fakeshadow commentedon Jan 2, 2026
Sorry I should have add more context to it as the bench site does not offer individual test based filter.
The
xitca-web [barebone]can be ignored as it'stokioonly and not realistic.xitca-webruns ontokio-uringandxitca-web [compio]runs oncompio. Both are almost identical implements on the http side.The postgres drivers are different. For
tokio-uringI took advantage of it's compat totokioecosystem and the driver is written in the way that can run ontokioentirely. Forcompiothe driver is rewritten for io_uring specifically. Therefore I consider them both are as efficient in their own way.From my observation
tokio-uringandcompioare at the same level of perf for the specific bench workload techempowers offers. That said I may not have either perform in their peak so there could be rooms left for either for improvement.fafhrd91 commentedon Jan 2, 2026
i have run compio in prod for sometime for large service. from my experience compio yields same results as tokio, no real advantages. i dont run it anymore, it may perform better in latest versions though
George-Miao commentedon Jan 2, 2026
I'm thinking about adding some performance-related documentation as suggested on reddit.
ararog commentedon Mar 4, 2026
I will resume efforts on make Deboa work with compio, this time based on cypher.
In other hand, my next project, a more robust web server, will be fully compio based.
George-Miao commentedon Mar 5, 2026
@ararog Sounds good! Feel free to post any results/problems. And also you can join our telegram group for a more timely respond.
ararog commentedon Mar 5, 2026
@George-Miao is it possible consider have a discord server too?
George-Miao commentedon Mar 5, 2026
@ararog We do have one: https://discord.gg/bGG8EF9v, just there's nobody except me in there. Feel free to join and chat!
0x676e67 commentedon Mar 16, 2026
This is my first time using compio, and I've just adapted it for btls (a fork of boring): https://github.com/0x676e67/btls/pulls. Everything seems to be working fine in testing.
Next, I plan to add compio support to wreq: 0x676e67/wreq#1097