How does Clojurescript compilation scales with CPU?

[Khalid Jebbari]

Hello,

We're using the parallel build option, and I noticed the difference in speed between my laptop and other laptops is basically proportional to the difference in speed of the CPUs (based on notebookcheck's benchmarks). I have 4C/8T 7700HQ CPU and my colleagues have a 8565U iirc (some have the 6600U). Mine is almost twice as fast in benchmarks, which is reflected in cljs compilation times.

So my question is how does the cljs compiler scale with regards to CPU? Core count? Single thread perf? All cores frequencies? Is it capped to some number of cores?

[Thomas Heller]

It depends on the namespaces used. In general a CLJS namespaces can only be compiled once all its dependencies have been compiled. So if those dependendencies can be compiled in parallel they will use multiple threads from a pool, which should keep all cores busy. In my experience a good balance between core count and core speed matters. If you have big namespaces that a lot of other namespaces depend on (eg. like cljs.core) then its compilation will "block" all other threads so its important it finishes fast (ie. fast cores). If you have lots a small namespaces that are mostly independent then you can get maximum parallelization (ie. many cores). 

I have a i7-8700K 6c and there are builds that aren't able to use all cores due to the namespace setup (few very large ones). Others happily use everything. Single core difference is gigantic to my previous CPU from a macbook pro 2016.

If you really really want to torture your CPU you can try https://github.com/mfikes/fifth-postulate or https://github.com/mfikes/coal-mine to compare.

HTH,

Thomas

[Khalid Jebbari]

Thanks for the detailed answer. Does it somewhat mean that splitting code into smaller namespaces can achieve faster compilation thanks to parallelization ?

--
Note that posts from new members are moderated - please be patient with your first post.
---
You received this message because you are subscribed to a topic in the Google Groups "ClojureScript" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/clojurescript/PwpVJNrF0Zc/unsubscribe.
To unsubscribe from this group and all its topics, send an email to clojurescrip...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/clojurescript/7a967f70-995e-49d3-9b25-a5b327689736%40googlegroups.com.

[Thomas Heller]

Depends on the size I'd say but in theory yes. Lots of things factor into the compilation times, even tiny namespaces can take a long time to compile if some macro just takes a long time to do its thing.

To unsubscribe from this group and all its topics, send an email to clojur...@googlegroups.com.

[Khalid Jebbari]

I may do some testing in the a close future and report here. I will either do it in our project or in a an open-source project that's "relevant" to show real-world numbers. Do you have an idea of such project? I think I will try to bench compilation time in advanced mode with a different number of cores activated by BIOS (I will soon have a 16C/32T machine, I hope the BIOS will allow me to deactivate some of them).

Khalid aka DjebbZ

@Dj3bbZ

To unsubscribe from this group and all its topics, send an email to clojurescrip...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/clojurescript/fe80b84a-4c23-4789-9b27-8f84c631d93d%40googlegroups.com.