2003-05-31 11:56:12

by ismail donmez

[permalink] [raw]
Subject: About 2.5.70-mm3

Hi Andrew,

I tried today 2.5.70-mm3 today and on an ext3 filesystem sytem startup
improved like 5-6 seconds same goes for X ( Compared to vanilla 2.5.70 ). Is
it because of "anticipatory scheduling elevator" or any other trick?

Regards,
/ismail


2003-05-31 20:08:35

by ismail donmez

[permalink] [raw]
Subject: Re: About 2.5.70-mm3

On Saturday 31 May 2003 20:44, you wrote:
> There's a little hack in there which speeds up the loading of executables:
> when someone does a mmap of a file with executable permissions the kernel
> will slurp it all into pagecache during the mmap. That tends to speed up
> program loading quite a lot, because the normal demand-loading produces
> quite poor I/O patterns.
Cool

>
> I had a vague feeling that this code wasn't working actually, and
> reimplemented it for -mm4.
You do not really feel it in X but on system startup its *quite* impressive.


Regards,
/ismail

2003-06-02 05:53:16

by Valdis Klētnieks

[permalink] [raw]
Subject: Re: About 2.5.70-mm3

On Sat, 31 May 2003 23:21:22 +0300, "ismail (cartman) donmez" said:
> On Saturday 31 May 2003 20:44, you wrote:
> > There's a little hack in there which speeds up the loading of executables:
> > when someone does a mmap of a file with executable permissions the kernel
> > will slurp it all into pagecache during the mmap. That tends to speed up
> > program loading quite a lot, because the normal demand-loading produces
> > quite poor I/O patterns.
> Cool

Yes, *majorly* cool. Between that tweak and the anticipatory scheduler,
it makes this laptop even more responsive than stock 2.5.70 was, and less
prone to hiccups - without -mm3, it was pretty easy to make xmms pause/skip
by suddenly hitting the disk with I/O. I've been trying to make it skip
for an hour, even doing things like launching an X application while running
an MH 'scan' command on a 7,000 item folder (which involves an open, read,
close for 7,000 separate files), and that doesn't phase it at all..

> > I had a vague feeling that this code wasn't working actually, and
> > reimplemented it for -mm4.
> You do not really feel it in X but on system startup its *quite* impressive.

It's noticeable on launching X apps on my laptop, and at boot time, I was
wondering if my init scripts were scrogged - [OK] after [OK] faster than
I'd seen before.

Amazing work, thanks Andrew (and all the contributors...)


Attachments:
(No filename) (226.00 B)