> > There are two problems which caused the install to run out of memory
> > on my 32Mb machine -
> > 1. The root (shm) filesystem defaults to half available RAM, ie. 16Mb.
> > If you set it to default to 24Mb, or even 20Mb, it would work (after
> > all installer components were unpacked it stood at 19.5Mb).
> > 2. The installer defaults to loading a whole bunch of unnecessary
> > modules. I know this is all about making it work smoothly on loads of
> > different hardware setups, but one solution would be to force the
> > debconf priority right down if it finds that main RAM <48Mb. With
> > careful selection of components, I suspect you could manage the
> > install on a 16Mb machine, let alone a 32Mb one. Nowadays, anyone
> > who's trying to install this on a 32Mb or lower machine probably knows
> > which hardware drivers they need ....
> Ok, I had not realized you were using beta1. This is fixed in the faily
> builds.
Great, apart from the fact that the bootfloppy.img file is *still* not included in the ISOs. It's not even kept in the same place as the daily ISO builds. Until you provide them together, it won't be unusual for people to be using the wrong version (I was using the 4 Jan ISO with the 9 Nov boot floppy!). Is there some deliberately policy of not including it??
> On December 15th we changed d-i to use tmpfs, with size=100mb. This
> should make it use as much memory as needed for d-i, up to 100 mb. A bit
> later, we made it free the boot ramdisk properly, which knocked several
> MB off of the space requirements. And like I said, I have tested it on
> machines artificially limited to 32 mb at boot time, and it works.
That all sounds excellent. What have you done about swapping - anything? I had to do it without any swapping in the end, because if I turned on a swap partition, even on a different disk, the "configure and mount partitions" step fell over as it tried to turn on a swap partition that was already in use. You might like to separate "turn on swap space" from "configure and mount install partitions" in order to avoid this. At the very least you could make the swapon code intelligent enough to avoid crashing out if a swap partition is already successfully in use as swap space. (Actually you need to do this whether you separate the two steps or not.)
Final installation report coming,
CC