[Frugalware-devel] i686 and PAE.

CSÉCSY László boobaa at frugalware.org
Sat Apr 23 21:32:19 CEST 2011


2011.04.23. 16:21:01-kor Krisztian VASAS írta:
> On 2011.04.23. 14:29, Miklos Vajna wrote:
> > On Fri, Apr 22, 2011 at 06:26:05PM -0500, James Buren<ryuo at frugalware.org>  
wrote:
> >> Would it be wise to enable PAE on i686 if it is not already done? Just a
> >> thought of something for 1.5.
> > 
> > What are your arguments for it?
> > 
> > Here are mines against it:
> > 
> > - it's slow
> 
> Yes, it _was_ slow, but now till 16GB there is no real slowdown
I don't know the current status of this topic; my latest first-hand experience 
is about half a year old (or even older). Anyway, I have been playing with 
this PAE dance on four different boxes, which gave me (at least) three 
different classes of results (I think based on the motherboard, but I'm not a 
hardware guy). I was trying to make use of the 4 GB RAM that was installed, 
which is not possible with a 32-bit kernel and without PAE – so I was trying 
the PAE-enabled 32-bit kernel on all the boxes.

1. Best result: PAE-enabled kernel enabled me to make use of all the 4 GB RAM, 
without any side effects (eg. slowdown).

2. Worst result: PAE-enabled kernel enabled me to make use of all the 4 GB 
RAM, but with terrible slowdowns (it was better to go with only 3.2 GB of the 
RAM actually available).

3. "No" result: I could still only use 3.2 GB of the RAM, without other 
effects.

Don't ask me the kernel version, nor mobo ID's; I can dig up the latter but I 
don't remember if they fell into this class or the other.

> > - it affects systems where PAE is not needed
Based on the above, it seems to be quite true: PAE-enabled kernel could help 
one user, but could make a different user's life hell in the same time.

> > - show me a system with 4G of RAM whose CPU is not 64bit-capable
Some googling could reveal the correct wording of a Linus' quote, which says 
something like that anything beyond 2 GB and still 32 bit is pure luck, in 
either way.

> Just think a bit: if there is a system that needs more cpu power and
> more ram, but the application does not support 64bit (probably a binary
> application), the PAE can be useful.
This is true – but I would run a 32-bit virtualized environment (or simply a 
32-bit chroot, which is possible, too) on a 64-bit box if I wanted to make use 
of all the RAM _and_ use a 32-bit only app (most probably a blob like skype).

Btw: skype is a weird animal: it's "64-bit" version needs multilib, which 
means it's a 32-bit app packaged into/for a(n already wicked) 64-bit 
environment. So to sum up my 2 pence, I vote against PAE, but vote for 
multilib (whatever that is in its deeps, as I'm newbie on this topic).
-- 
Bye, Boobaa (CSÉCSY László)
http://csecsy.hu


More information about the Frugalware-devel mailing list