Minor speed tweak tip

[Shern, Benjamin J]

I definitely agree that this is not useful for lots of (most?) people and
probably not worthy of any special tuning parameters.

One of the problems is that my results are empirical and the actual
explanation of the speed difference is not known for sure.

I'm wondering if some or all of the timing differences I saw can be
attributed to the fact that my tmpfs has ~1,000 inodes and my root fs has
~200,000 inodes...

Regarding swapping the entire ramfs; I have yet to see my system really
thrash even under heavy load.  I  believe the linux MM has undergone a lot
of improvements in the 2.4+ kernels.


> -----Original Message-----
> From: Auke Kok [mailto:auke.kok at planet.nl] 
> Sent: February 10, 2003 4:29 PM
> To: This is the primary mailing list for Lunar Linux!
> Subject: Re: Minor speed tweak tip
> 
> 
> On Mon, 2003-02-10 at 23:20, Shern, Benjamin J wrote:
> > For those of you with a decent size chunk of RAM:
> > I've benchmarked 5-10% improvement in compile times by 
> caching /usr/bin into
> > memory.  The best way I've found to do this is:
> 
> just some brief thoughts...
> 
> In a way you've found that linux is bad a cacheing binaries. 
> Allocating
> memory away from the system cache actually sounds like a bad 
> idea right?
> In all people with small amounts of memory are helped with LOW MEMORY
> usage by system stuff and not too-big caches, because they take away
> space that the kernel could otherwise allocate for something more
> usefull.
> 
> All in all it's not really usefull for most people, and yes, 
> the kernel
> fs cacheing should have been a bit smarter.
> 
> another side not is that your whole ramfs will be swapped out if the
> kernel requires it, killing your performance of course when you peak
> memory usage once every so often.
> 
> sofar
> 
> > 
> > mkdir /usr/bin_ram
> > mount -t tmpfs tmpfs /usr/bin_ram
> > cp -a /usr/bin/* /usr/bin_ram
> > 
> > Edit /etc/profile and put /usr/bin_ram before /usr/bin.
> > 
> > Only caveat is that you're mindful to re-mirror the tmpfs 
> after removing
> > packages.
> > 
> > Even on systems with less memory I'm guessing some 
> performance could be
> > added during compiling by caching individual binaries such 
> as gcc, binutils,
> > make, install etc...
> > 
> > regards,
> > Ben
> > 
> > 
> > Curious if anyone else has tested this and seem similar or 
> better results...
> > _______________________________________________
> > Lunar mailing list
> > Lunar at lunar-linux.org
> > http://lunar-linux.org/mailman/listinfo/lunar
> -- 
> Auke Kok <auke.kok at planet.nl>
> 
> _______________________________________________
> Lunar mailing list
> Lunar at lunar-linux.org
> http://lunar-linux.org/mailman/listinfo/lunar
>