Distribuirano kompajliranje na vise masina

• 3Ovo se svidja korisnicima: Rogi, mcrule, ThePhilosopher
  
  Registruj se da bi pohvalio/la poruku!

Napisano: 03 Feb 2011 12:19

Posle mozda malo zbunjujuceg naziva teme sledi odlican tekst o tome kako ubrzati kompajliranje tako sto se proces kompajliranja "distribuira" na vise masina koristeci distcc, i ccache. U primeru se koriste samo dve Sun SPARC masine i NetBSD sparc64, ali je proces totalno isti i za vise masina.

Citat:Distributed Compilation, Part I


Introduction

I have several old Sun machines, which are all still working fine, but partly lack performance of modern computers and/or RAM.

I usually install NetBSD on my computers, especially on those Sun-boxes. The advantage is that every machine has the same "look and feel", I know exactly what to do, etc. The downside is that for this kind of platform I have to use pkgsrc and build everything by myself, as the software I like to use is oftentimes not available as a pre-built binary, and besides that, I usually use special CFLAGS to tweak the performance of the software being built.

So, sitting in front of my new favorite toy, my Ultra-60, I was thinking about how I could probably speed up compilation timings. Looking around, and seeing several spare machines that were ready installed with a NetBSD 5.1, but switched off and doing nothing, I had the idea to create a compilation-cluster. I knew that there was already software available to achieve ths, called "distcc". All I needed to do, was to install it on every machine, read the man-pages a bit, and give it try.

...
http://blog.nifelheim.info/tech/2011/02/02/distributed-compilation-part-i/

Dopuna: 17 Avg 2011 10:49

Pojavio se drugi deo, autor je dodao jos jednu masinu, ali sada i386 (prosle dve su sun sparc), demonstrirajuci kako se cross-compiling moze izvesti bez problema:

Citat:Distributed Compilation, Part II

As I wrote in my previous Post, "Distributed Compilation, Part I", the so far gained speed up of compilation times was significant, but it still took long.
The most obvious solution was to add more hosts to the list, but which hosts should that be? The Ultra-5 of course not, and all other Sun computers were slower and less well equipped (with the exception of the Ultra Enterprise 2), and each of them would add a significant amount to our power-consumption. There is however panther. The sleeping power of the dual-cored Atom 330 with enabled hyperthreading is well known from compiling NetBSD, which, starting from scratch until completion of a ready-to-burn ISO-image including Xorg, usually takes about 3.5-4.0 hours depending on the target-architecture and whether the toolstack has to be created first.
Panther would not add additional power-consumption (at least not significantly) to the overall-power-consumption, and was idle most of the time. The only problems were these two:
is cross-compilation possible at all?
how to get the cross-compiler working?
...

[Link mogu videti samo ulogovani korisnici]