Benchmarks

[GuacheSuede]

Hi Leandro (@lpereira),

I ran some benchmarks on LWAN and realized subsequent requests appear to be much faster than the first.
Which part of the codes does this "caching" ?

Thank You

[lpereira]

If you're benchmarking file serving, it's partially in lwan-cache and lwan-mod-serve-files. If you're doing that, then there's no caching.

…

On Wed, Nov 25, 2020, 10:36 Justin ***@***.***> wrote: Hi Leandro, I ran some benchmarks on LWAN and realized subsequent requests appear to be much faster than the first. Which part of the codes does this "caching" ? Thank You — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <https://github.com/lpereira/lwan/issues/297>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAADVGJQH7SHYUXEPXZQNQLSRVE3DANCNFSM4UCYF5UA> .

[GuacheSuede]

I see, what about Normal API calls ? Would it be due to object initialization ? I noticed that if left running and not called for a while, when the web server is called again for the first time, there is a slight drop in latency with the subsequnet requests being faster once again.

Eg. START 9ms 3ms 3ms 3ms 3ms LONG BREAK 9ms 3ms 3ms 3ms

[lpereira]

Is the pause necessary to observe this behavior? Because if you do use keep-alive connections, a lot of memory will be kept around (and reallocated as soon as the connection is closed). This includes stack space for coroutines, response buffers, etc.

…

On Wed, Nov 25, 2020, 17:18 Justin ***@***.***> wrote: I see, what about Normal API calls ? Would it be due to object initialization ? I noticed that if left running and not called for a while, when the web server is called again for the first time, there is a slight drop in latency with the subsequnet requests being faster once again. Eg. START 9ms 3ms 3ms 3ms 3ms LONG BREAK 9ms 3ms 3ms 3ms — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <https://github.com/lpereira/lwan/issues/297#issuecomment-734018767>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAADVGMTFEIOLYJBM5ZYDB3SRWUFXANCNFSM4UCYF5UA> .

[GuacheSuede]

I see, Thanks for the reply, think that clears that up.

Was studying Lwan's code, out of curiosity, could u explain what concurrency model does lwan use ?

[lpereira]

While a bit outdated, this might help you understand how it works: https://tia.mat.br/posts/2014/10/06/life_of_a_http_request.html The model is closer to a reactor model (non-blocking sockets, multiple threads each one blocking on epoll(), etc.) than a proactor but I don't really know for sure. Never really studied these things.

…

On Fri, Nov 27, 2020 at 8:47 AM Justin ***@***.***> wrote: I see, Thanks for the reply, think that clears that up. Was studying Lwan's code, out of curiosity, what concurrency model does lwan use (proactor/reactor etc..) ? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <https://github.com/lpereira/lwan/issues/297#issuecomment-734718623>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAADVGPOOLNNDFZJMBAZJHDSR5RQLANCNFSM4UCYF5UA> .

-- Leandro

[GuacheSuede]

Thank You for the clarification.

Is it possible to modify the url_map run time ? via referencing the map as a pointer ?

[lpereira]

Possible? Yes. Recommended? No.

…

On Fri, Nov 27, 2020, 10:08 Justin ***@***.***> wrote: Thank You for the clarification. Is it possible to modify the url_map run time ? via referencing the map as a pointer ? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <https://github.com/lpereira/lwan/issues/297#issuecomment-734941569>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAADVGM6LCINLOIY76JLZATSR7TI5ANCNFSM4UCYF5UA> .

[GuacheSuede]

I see, is there a performance penalty for doing that ?

[lpereira]

No -- it might make Lwan use deallocated memory if a request comes in. At best this will crash.

…

On Sat, Nov 28, 2020 at 4:00 PM Justin ***@***.***> wrote: I see, is there a performance penalty for doing that ? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <https://github.com/lpereira/lwan/issues/297#issuecomment-735248753>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAADVGIS4ZJVRBKKM45GHDDSSENAPANCNFSM4UCYF5UA> .

-- Leandro