Re: Unexec'ing Scwm [was Re: Proposal for a Guile binary file format]

[thi <ttn at mingle dot glug dot org>]

Greg J. Badros writes:

 > >  scheme-code directly:   5 ms
 > >  using shared lib:      22         [...]
 > 
 > Wait, so you're writing that it takes between 3 and 4 times longer to
 > load the shared libraries than just the plain scheme code?  Scwm is
 > plenty fast executing Scheme code (and when it's not fast enough, we
 > can just write a new primitive)--- it's plenty slow *loading* and
 > that's what I'm hoping to speed up.
 > 
 > Please let me know that I'm misinterpreting your performance numbers.

i don't think you are misinterpreting the numbers.  however, the numbers
cannot claim to be "representative" and so extrapolating from them is
probably unwise.  also, there may be other one-time costs that i'm not
aware of that dominate the loading performance.

i'm not familiar w/ the memoization process and its specific timing, but
my guess (to explain the numbers) would be that loading directly defers
memoization -- you pay for it later -- whereas loading a shared library
takes care of all that immediately.  but this is just a guess; hopefully
someone clueful will chime in w/ insight...

thi