Preloading [WAS: Re: SuSE Project SUPER]

[Andrew Farris]

Linus Walleij wrote:
> Preload then uses readahead(2) or fadvice(2) to do the actual preloading.
> 
> Then preload is triggered from /etc/init.d scripts prepended with 
> ionice(1) like this:
> 
> /usr/bin/ionice -n2 /sbin/preload /etc/preload.d/gdm
> 
> Then as can be seen there are some app profiles in /etc/preload.d that 
> configure depending on used desktop what to preload, e.g. gdm/kdm, 
> Khelpcenter, OpenOffice, Firefox, Gimp (why?), etc.

Incidentally, I think one of the main reasons the idea hasn't had much 
discussion is that question of what applications are so crucial to preload like 
this (gimp? laughable).  Also, once an application is open its shared libraries 
are cached for later, so for machines that stay on all day this has almost no 
real benefit.

> All code and scripts are heavily SuSEified.
> 
> But has this really been under our radar for all this time? The entire 
> concept is like 2-3 years old, surely it must have been discussed 
> before? (Not that I heard it though...)

I've seen discussion of this concept at least, maybe not that particular 
implementation strategy.  The idea should provide some better performance from 
the user perspective... but the fact is the real wall time of getting your 
machine cold booted, logged in, and an application fired up and usable should 
not be much different overall.  The real differences would be seen if you are 
not sitting at the machine logging in asap for instance (the machine has nothing 
to do until you log in, whereas with that preloading it might start loading up 
firefox for you).

If you put this in the distro, you need good high-level configuration tools so 
the user can choose what to preload or not, so you're not slowing the machine 
unnecessarily, or packing all the free memory with libraries the user doesn't 
intend to use that day/boot/login just to later unload it.

I wouldn't even want something learning my most used applications, just 
something to configure what will get loaded and what won't.

-- 
Andrew Farris <lordmorgul at gmail.com> <ajfarris at gmail.com>
  gpg 0xC99B1DF3 fingerprint CDEC 6FAD BA27 40DF 707E A2E0 F0F6 E622 C99B 1DF3
No one now has, and no one will ever again get, the big picture. - Daniel Geer
----                                                                       ----