> Miquel van Smoorenburg writes:
> > In article <Pine.LNX.3.95.971119095638.28363A-100000@chaos.analogic.com>,
> > Richard B. Johnson <root@chaos.analogic.com> wrote:
> > >On Thu, 20 Nov 1997, Richard Gooch wrote:
> > >
> > >> Hi, all. I have a perl script (latex2html, actually) which reads
> > >> lots of small input files and then writes to some output file(s). I
> > >> notice that with 2.1.65 my disc head is moving quite a bit, which
> > >> didn't happen with 2.1.62. This happens even if all the input files
> > >> are in the cache. With 2.1.62 this processing was almost silent, now
> > >> with 2.1.65 there is heaps of head movement (and the operation seems
> > >> slower).
> > >[SNIPPED]
> > >
> > >Get the source code for init. Search for "sync()". Comment it out
> > >and re-compile.
> >
> > That might help, yes. Note that sysvinit-2.72 (the latest) already does away
> > with the sync() in the main loop.
>
> Yes, but... what's changed from 2.1.62 to 2.1.65? I didn't get this
> behaviour with 2.1.62.
>
> Regards,
>
> Richard....
>
I think there are now fewer entries left in the dcache. Originally
entries were left until the machine was basically out of ram. This
improved performance when there were a lot of directory snippets from all
over being parsed and/or executed. However, big programs had problems
getting RAM.
Now I see a lot of pruning going on to keep some free RAM. Scripts may
suffer a bit, but programs may fare better.
With old versions of init, every time a child exited to init, the file-
system was synched. This made for a lot of disc activity when scripts
were executed as `sh do_something`, trapped by init and not bash.
Cheers,
Dick Johnson
Richard B. Johnson
Project Engineer
Analogic Corporation
Penguin : Linux version 2.1.65 on an i586 machine (66.15 BogoMips).
Warning : It's hard to remain at the trailing edge of technology.