| 1 |
On 2010-12-03, Alan McKinnon <alan.mckinnon@×××××.com> wrote: |
| 2 |
> Apparently, though unproven, at 00:37 on Saturday 04 December 2010, Grant |
| 3 |
> Edwards did opine thusly: |
| 4 |
> |
| 5 |
>> On 2010-12-03, David W Noon <dwnoon@××××××××.com> wrote: |
| 6 |
>> > What you are seeing is a lot of glibc routines being included by the |
| 7 |
>> > linkage editor. These handle all sorts of conditions that will |
| 8 |
>> > likely never occur in your program. |
| 9 |
>> > |
| 10 |
>> > Try using a smaller C library, like uclibc or klibc. They might not |
| 11 |
>> > work as well, but they will give you a smaller executable. |
| 12 |
>> > |
| 13 |
>> > Alternatively, try rewriting your code in assembler. |
| 14 |
>> |
| 15 |
>> For various reasons (which I doubt anybody cares about), gritting my |
| 16 |
>> teeth and living with the 520K per application looks like a more |
| 17 |
>> practical solution that either using assembler or a different libc. |
| 18 |
>> |
| 19 |
>> In practice, I'm sure nobody but me will ever even notice (or care |
| 20 |
>> even if they did notice) the wasted 2MB on a 25MB liveCD. But it will |
| 21 |
>> still bug me. :/ |
| 22 |
> |
| 23 |
> That's 8% of your space resources. |
| 24 |
|
| 25 |
It's closer to 0.5%. My space resources are 700MB on the CD and |
| 26 |
probably at least 512MB of RAM. Right now I'm using up about 22MB of |
| 27 |
the space on the CD, and less than that in RAM. By the time I'm done, |
| 28 |
it will probably be around 25MB. Anything that boots from a single CD |
| 29 |
in under 15 seconds and runs in 512MB of RAM or more will be fine. The |
| 30 |
difference between 23MB and 25MB really won't be noticable. |
| 31 |
|
| 32 |
> Many more than just you will notice and care and whinge loudly. |
| 33 |
> Probably including me. |
| 34 |
|
| 35 |
Well, I think you'd find it pretty useless no matter how small/fast it |
| 36 |
was. The only thing it's able to do is production test and |
| 37 |
diagnostics for a family of PCI boards I support. |
| 38 |
|
| 39 |
> It might be worth the effort to switch to a libc designed for the |
| 40 |
> sort of task you want to accomplish. I assume you already made the |
| 41 |
> effort with busybox or similar, it's much the same viewpoint. |
| 42 |
|
| 43 |
Sort of. I'm using busybox, but the "base" of my CD is generated by a |
| 44 |
shellscript that strips down a systemrescuecd ISO image to the bare |
| 45 |
essesntials (basically just what's required to run a 7MB initrd image |
| 46 |
containing busybox). Then I add my three or four static applications |
| 47 |
to that. We used to build our own CD from scratch, but that build |
| 48 |
process was fragile and hard to maintain. |
| 49 |
|
| 50 |
The CD doesn't even have network support or drivers for any block |
| 51 |
devices except a ramdisk (no SCSI, SATA, ATA, and so on). The only |
| 52 |
filesystem it supports is cramfs. I ripped out USB support also, but |
| 53 |
then had to put enough of it back in for the USB keyboards on some of |
| 54 |
the machines where it'll be used. |
| 55 |
|
| 56 |
It'll still bug me that those static apps are 90% bloat, but it's just |
| 57 |
not worth the effort to switch to uClibc or mess with putting my own |
| 58 |
dynamic libraries onto the CD. |
| 59 |
|
| 60 |
-- |
| 61 |
Grant |
Archives