compiler / glibc optimization

[Albert Cahalan]

On Jan 25, 2008 4:29 AM, Wade Brainerd <wadetb at gmail.com> wrote:

>> People with the hardware should develop exclusively on the hardware.
>
> This made me smile at the thought of the "OLPC Project grinds to
> a sudden halt, no commits in weeks" headline.

You think? If so, there is something wrong already. Lack of
dogfooding is why Sugar no longer runs on 128 MB hardware.
People just don't care much about bad performance unless they
are forced to live with it.

Problems get fixed mighty quick when they are in your way.

Sure enough, we're back to having activities die. The base OS
has grown quite a bit. Without dogfooding, it will continue to
grow until the system is mostly unusable like the B2 hardware.
The OS ought to be functional in well under 32 MB, with most
of the activities workable in 64 MB.

Just a decade ago, XO-level performance was damn good.
Go back 17 years, to when Linus was hacking on a 386SX-16,
and the XO level of performance was pretty much unobtainable
at any price. (government labs excepted... maybe)

Ever wonder why Linux supports swap partitions? Linus had 4 MB
of RAM, which was a luxury. Many people only had 2 MB. Adding
a 2 MB swap partition allowed them to run gcc.

Really, the XO is fully capable of being a primary development
workstation. This won't work for long without dogfooding.
Remember that the kids will have nothing else. This is it for them.
They can't build on a Mac with a pair of quad-core Core 2 Xeons
and 4 GB of RAM. It's the XO or nothing.