This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: How would *you* use an intern?
- To: DJ Delorie <dj at redhat dot com>
- Subject: Re: How would *you* use an intern?
- From: "Zack Weinberg" <zackw at Stanford dot EDU>
- Date: Thu, 31 May 2001 00:43:30 -0700
- Cc: pfeifer at dbai dot tuwien dot ac dot at, shebs at apple dot com, gcc at gcc dot gnu dot org
On Thu, May 31, 2001 at 03:29:46AM -0400, DJ Delorie wrote:
>
> > Linking files residing on an NFS-Server often gives me a load of < 5%
> > with a most significant amount of iowait, even on a fast network.
> >
> > I assume this could be improved by de-serializing the link process?
>
> Yeah. The linking process is highly parallelizable. Loading objects
> is, resolving symbols is, laying them out in memory isn't, relocating
> is, writing out the executable isn't (well, might be).
Ah, I didn't know that. Thanks.
The last time I looked at libbfd bottlenecks in any detail was 1997 or
so, and the main trouble for the case I was interested in was doing
lots of seeks and small writes via stdio. I remember thinking that
for most things it'd be more efficient to dump everything into a big
memory block and then blit that to disk all at once with low level I/O
primitives.
Mind, that wasn't ld I was looking at, that was *ar.* Why ar has to
do seeks is completely beyond me.
--
zw "But that means every licensed realtor in America is a creature from outer
space!" I shouted. "What are we going to do about it?"
"Well," said Osgood Sigerson, the world's greatest detective, "I suppose
we'll just have to live with it."
-- Daniel Pinkwater, _The Snarkout Boys and the Avocado of Death_