Results of actual compile printk format compression

Results of actual compile printk format compression

[Timothy Miller]

Just a quick note...

Although my experiments with kernel printk format string compression 
have reported estimated shrinkage, this is the first time I have been 
able to compile a whole kernel with the compression filter.

These results come from doing an allyesconfig of 2.5.68 and then weeding 
out anything that didn't build.  One program extracts strings from 
preprocessor output, a second program determines how the strings will be 
encoded, and the third makes substitutions during a kernel compile.

The uncompressed compile resulted in a kernel image of 24011892 bytes. 
The resulting image with format strings compressed is 23904708 bytes 
which is a shrinkage of 107184 bytes.  Subtracting out an estimate of 3K 
for the dictionary and necessary modifications to printk, that results 
in a reduction of something like 104112 which is 4% of the original 
kernel size.

That may not seem like a lot, but if you consider only the printk 
strings themselves, they are compressed to less than 50% of their 
original size (counting the dictionary but not printk code mods).

So, I ask... is this a useful savings?  Is there any chance anyone would 
bother to increase their compile time by a factor of 5 in order to shave 
off 4% or 100k bytes?

(Not to mention that allyesconfig is a very unrealistic scenario.)

Re: Results of actual compile printk format compression

[Felipe Alfaro Solana]

On Sat, 2003-06-07 at 00:52, Timothy Miller wrote:
> Just a quick note...
> 
> Although my experiments with kernel printk format string compression 
> have reported estimated shrinkage, this is the first time I have been 
> able to compile a whole kernel with the compression filter.
> 
> These results come from doing an allyesconfig of 2.5.68 and then weeding 
> out anything that didn't build.  One program extracts strings from 
> preprocessor output, a second program determines how the strings will be 
> encoded, and the third makes substitutions during a kernel compile.
> 
> The uncompressed compile resulted in a kernel image of 24011892 bytes. 
> The resulting image with format strings compressed is 23904708 bytes 
> which is a shrinkage of 107184 bytes.  Subtracting out an estimate of 3K 
> for the dictionary and necessary modifications to printk, that results 
> in a reduction of something like 104112 which is 4% of the original 
> kernel size.
> 
> That may not seem like a lot, but if you consider only the printk 
> strings themselves, they are compressed to less than 50% of their 
> original size (counting the dictionary but not printk code mods).
> 
> So, I ask... is this a useful savings?  Is there any chance anyone would 
> bother to increase their compile time by a factor of 5 in order to shave 
> off 4% or 100k bytes?
> 
> (Not to mention that allyesconfig is a very unrealistic scenario.)

Sincerely, with machines ranging 512MB of memory, 100KB of memory saving
is just so little, that factoring kernel compilation by 5 is something
I'm not willing to take.

However, it could be useful to include that feature for memory
constrained systems, like embedded ones, and allow enabling it at
compile/configure time.