weird science experiments building on low-spec hardware

Forum rules
We believe in Hello and Thank You.
pidsley
Hermit
Posts: 2539
Joined: Wed Oct 17, 2012 12:31 pm

weird science experiments building on low-spec hardware

Unread post by pidsley » Sun Jul 20, 2014 7:07 pm

kexolino on IRC asked me about running source-based distros on low-spec hardware, and then someone on CRUX IRC mentioned he was running CRUX on an Atom 510, so of course I had to try it on my Atom.

It is possible to install and run CRUX on an Atom 410. This is a 1.6 GHz single-core mulithreaded processor (so it looks like it has two cores, but really only time-shares one core). It takes approximately 35 minutes to build a kernel on this machine, and the first system update took around a half hour. Updating something big like glibc would probably take several hours, and I would not even try build something big like firefox or libreoffice. This is also on my open fan-cooled atom board; I would not recommend trying this on an Atom netbook; it puts a heavy load on the processor and I would be very worried about overheating the netbook.

Image

I also have CRUX running on a 2.0 GHz single-core sempron machine. Builds on this machine take almost as long as on the Atom, but the sempron never even gets close to overheating.

I decided to run some speed comparisons, using compiling a 3.15.5 kernel with the same config file on several machines. Here are the numbers:

Code: Select all

2.33 GHz core2duo   7 minutes
2.5 GHz athlon x2  10 minutes
2.0 GHz sempron    24 minutes
1.6 GHz atom       35 minutes
2.0 Ghz celeron    46 minutes
The Celeron is 32-bit, so I built a 32-bit kernel there, just for time comparison. This is the machine where I used to run 32-bit Gentoo, and updates would often take all night.

Romster on CRUX IRC mentioned using distcc for big compiles, so that was my next experiment. distcc allows you to share a compile job among two or more machines, and can make a large or small difference, depending on the linked machines, and on the job itself.

I used the same 3.15.5 kernel build as a test, and here are some numbers:

Code: Select all

core2+sempron         5m25s
core2+athlon          4m14s
core2+athlon+sempron  4m7s
core2+athlon+sempron+atom  3m46s

core2+atom   5m44s
I did the last test just to see how much difference adding just the Atom would make.

Romster told me that all the distcc machines have to be using the same version of CRUX, but I'm not sure if this means they have to be using the same version of distcc and gcc, or if the reasons are deeper than that. At any rate, I think the next time I have a big update (or even a kernel build) I will try distcc with the Athlon and Core2 working together.

references:
http://www.linux-mag.com/id/4874/
https://code.google.com/p/distcc/

User avatar
franksinistra
Ivana Fukalot
Posts: 1093
Joined: Mon Jan 27, 2014 2:03 am
Location: 印尼国

Re: weird science experiments building on low-spec hardware

Unread post by franksinistra » Sun Jul 20, 2014 7:25 pm

wow!!!
i'll try this on my netbook (atom n270 netbook) just for fun
rice no more.

User avatar
GekkoP
Emacs Sancho Panza
Posts: 5878
Joined: Tue Sep 03, 2013 7:05 am

Re: weird science experiments building on low-spec hardware

Unread post by GekkoP » Mon Jul 21, 2014 8:26 am

Very interesting experiments, Pidsley.

Post Reply