[sldev] Re: shared objects output during standalone build

[Dale Glass]

On Tue, Jul 17, 2007 at 09:47:42AM -0400, Jason Giglio wrote:
> I've found that on my Quad Core single chip, it's best to just stay with 
> the actual number of cores you have, i.e. -j 4 is fastest.
> 
> This could be because of disk bottlenecks.
My own experiments indicate that build speed depends very heavily on the
amount of RAM available. Enough RAM will make disk speed completely
irrelevant. If you have too little though, the build process will result
in evicting useful data from the cache, and making the disk a lot more
relevant.

I've seen g++ use somewhere about 300-500MB per process while building.

I'd estimate the amount of RAM needed to compile SL optimally to be
somewhere about:

(400MB * num_processes) + 400MB + amount_used_by_applications

The +400MB is the approximate amount of intermediate results the build
generates. In the ideal case, all that remains cached in RAM by the time
the build is done.

So a 8 process build would require a 64 bit box (32 bit has a practical
RAM limit of around 3GB) with somewhere about 5GB, assuming 1GB for the
IDE and such. Add another GB or so if you're going to run SL at the same
time (although that'll take one core of course)

Here I do a full build of SL in approximately 15 minutes, on an
Athlon 64 X2 5200+ with 4GB RAM, running in 64 bit mode, using -j2.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://lists.secondlife.com/pipermail/sldev/attachments/20070717/904c4f79/attachment-0001.pgp