This is the mail archive of the
libc-hacker@cygnus.com
mailing list for the glibc project.
Re: comment on today's changes
- To: Geoff Keating <geoffk@ozemail.com.au>
- Subject: Re: comment on today's changes
- From: Ulrich Drepper <drepper@cygnus.com>
- Date: 01 Jan 1999 22:17:40 -0800
- Cc: libc-hacker@cygnus.com
- References: <r267awy9om.fsf@happy.cygnus.com> <199812291255.XAA02269@geoffk.wattle.id.au> <r2sody1j7b.fsf@happy.cygnus.com> <199812310641.RAA21273@geoffk.wattle.id.au>
- Reply-To: drepper@cygnus.com (Ulrich Drepper)
Geoff Keating <geoffk@ozemail.com.au> writes:
> > It does not matter for the case I was interested in. It's wall clock
> > time. If we would have a method which wouldn't have to touch many
> > pages the startup time is reduces. And this is the goal.
>
> The program, or glibc, may touch the pages as it starts up anyway...
The relocation are write accesses thus forcing the copy-on-write. The
idea is to prevent this and so, if the pages are alreday in the
filesystem cache, no extra costs exist.
Anyhow, the whole testing is not meant to be accurate. I just had to
convince myself that the copy-on-write actions take quite a lot of
time. When Richard and I talked about this we both were not sure it
is worth it. We know why SGI has implemented Quickstart (because MIPS
ELF does not allow lazy relocations) but at the same time Sun hasn't
implemented this. There are still some costs hidden in the processing
of the conflict list but whether this leads to major problems can only
be found out by implementing it.
--
---------------. drepper at gnu.org ,-. 1325 Chesapeake Terrace
Ulrich Drepper \ ,-------------------' \ Sunnyvale, CA 94089 USA
Cygnus Solutions `--' drepper at cygnus.com `------------------------