Slacko 5.4 PAE vs NON-PAE comparative Analysis

Under development: PCMCIA, wireless, etc.
Post Reply
Message
Author
User avatar
Q5sys
Posts: 1105
Joined: Thu 11 Dec 2008, 19:49
Contact:

Slacko 5.4 PAE vs NON-PAE comparative Analysis

#1 Post by Q5sys »

From the PAE discussion that's brewing...
Q5sys wrote: I was curious as to the results everyone here is speaking about. I would be interested in seeing the 'results' that people are basing their opinions on. Because the only results I have seen before show that PAE isnt quite as fast as non-PAE.
So i decided to spend some free time today (which I happen to have) and compare the newest Slacko 5.4 on 4 machines. I burned the PAE and the non-PAE ISOs, and booted them.
Each computer was tested the same way. Boot the system from the CD, load the OS into ram, and run Hard-info.
On Modern Hardware the difference is minimal, aside from a very interesting result on a a quad core i5 with 16gb ram. This system was fully able to utilize the benefits of PAE, however did so with an unexpected side effect. Obviously on an modern system with more than 4Gb ram, a user should be using a 64bit release.

The worst result as I expected was on older hardware, where ram is minimal and is precious. The system had enough ram to boot and as such the ISO loaded into ram. The extra memory used for PAE, which is required, impacts older minimal hardware more than modern hardware. On older hardware running PAE is a severe drain of precious resources, and as such should NOT be encouraged.

Overall the results are mixed. On Modern hardware PAE is slower than non-PAE in some tasks but not others. However on older hardware it was slower across the board, and on older hardware a 2% performance decrease is much more noticeable in the user experience than on a modern system.



Toshiba 7200 Cte Laptop
Total RAM on PAE is 2% less than non-PAE.
Free RAM on PAE is 36% less than non-PAE.
Benchmarks -
Blowfish - PAE is 2% slower than non-PAE
CryptoHash - PAE is 1% slower than non-PAE
Fibonacci - PAE is 1.5% slower than non-PAE
N-Queens - PAE is 1% slower than non-PAE
FFT - PAE is 2.7% slower than non-PAE
RayTracing - PAE is 2.3% slower than non-PAE

Acer Aspire One D257 Netbook
Total RAM on PAE is 1% less than non-PAE
Free RAM on PAE is 1% less than non-PAE
Benchmarks -
Blowfish - PAE is <1% slower than non-PAE
CryptoHash - PAE is 2% faster than non-PAE
Fibonacci - PAE is <1% slower than non-PAE
N-Queens - PAE is 2% faster than non-PAE
FFT - PAE is 1% slower than non-PAE
Raytracing - PAE is 2% faster than non-PAE

Lenovo Y510 with 4gb ram
Total RAM on PAE is 1% less than non-PAE
Free RAM on PAE is 3% less than non-PAE
Benchmarks -
Blowfish - PAE is 1% faster than non-PAE
CryptoHash - PAE is 1% slower than non-PAE
Fibonacci - PAE is 7% faster than non-PAE
N-Queens - PAE is <1% slower than non-PAE
FFT - crashes hard-info - so no results
Raytracing - crashes hard-info when run - so no results

Custom Built Intel i5 system with 16gb ram
Total RAM on PAE is 530% greater than non-PAE (what we would expect)
Free RAM on PAE is 603% greater than non-PAE (what we would expect)
Used RAM on PAE is 1% higher than non-PAE (what we would expect due to how PAE works)
Benchmarks -
IMPORTANT DISCLAIMER: PAE system @ 2800Mhz compared to non-PAE running @ 1200Mhz
Blowfish - PAE is <1% faster than non-PAE
Cryptohash - PAE is 9% slower than non-PAE
Fibonacci - PAE is <1% slower than non-PAE
N-Queens - PAE is <1% slower than non-PAE
FFT - PAE is <1% faster than non-PAE
Raytracing - PAE is <1% faster than non-PAE
This is not what I expected. On this sytem I expected PAE to be much faster than non-PAE. The fact that PAE is only marginally faster than non-PAE when run at 233% processor speed is very suprising to me. Moreso considering that when Cryptohash on PAE was running the CPU at 2800Mhz it was 9% slower than the non-PAE running Cryptohash with CPU at 1200Mhz.
(I preformed this test 3 more times and got the same results, so its not a random result)



LINKS:
http://q5sys.info/pae/tests/hardinfo_re ... 00cte.html
http://q5sys.info/pae/tests/hardinfo_re ... 00cte.html
http://q5sys.info/pae/tests/hardreports ... 200cte.png
http://q5sys.info/pae/tests/hardreports ... 0cte-2.png
http://q5sys.info/pae/tests/hardreports ... 0cte-3.png

http://q5sys.info/pae/tests/hardinfo_re ... -D257.html
http://q5sys.info/pae/tests/hardinfo_re ... -D257.html
http://q5sys.info/pae/tests/hardreports ... e_D257.png
http://q5sys.info/pae/tests/hardreports ... D257-2.png
http://q5sys.info/pae/tests/hardreports ... D257-3.png

http://q5sys.info/pae/tests/hardinfo_re ... -t510.html
http://q5sys.info/pae/tests/hardinfo_re ... -t510.html
http://q5sys.info/pae/tests/hardreports-Lenovo_Y510.png
http://q5sys.info/pae/tests/hardreports ... Y510-2.png
http://q5sys.info/pae/tests/hardreports ... Y510-3.png

http://q5sys.info/pae/tests/hardinfo_re ... _16gb.html
http://q5sys.info/pae/tests/hardinfo_re ... _16gb.html
http://q5sys.info/pae/tests/hardreports-i5_16gb.png
http://q5sys.info/pae/tests/hardreports-i5_16gb-2.png
http://q5sys.info/pae/tests/hardreports-i5_16gb-3.png

User avatar
mikeslr
Posts: 3890
Joined: Mon 16 Jun 2008, 21:20
Location: 500 seconds from Sol

Thanks for running the tests

#2 Post by mikeslr »

Thanks, Q5sys, for running the tests. I can appreciate how much time they took, having recently run compartive tests on pets vs. sfses vs. program folders.

About two years ago, when PAE kernels began to show up, I did a google search and reported pretty much the same thing based on a professional lab's tests under, albeit I think, Fedora. Since then, whenever the opportunity arose, I argued that Devs should be developing two species of Puppies: non-PAE for 32-Bit, PAE for 64-bit. Regretfully, I suspect I've been preaching heresy to True Believers: neither evidence nor reason matters.Or perhaps, as I've also said, Devs are likely to acquire modern, powerful, computers and create the best system for those. The thought that a non-PAE Pup may breath new life into hundreds of thousands of computers, otherwise slated for landfills, that may be of use in the 3rd World, or to most of the people now economically depressed in the 1st --in short, that Pups actually have a social value-- no longer appears to be important.

mikesLr

User avatar
Q5sys
Posts: 1105
Joined: Thu 11 Dec 2008, 19:49
Contact:

Re: Thanks for running the tests

#3 Post by Q5sys »

mikeslr wrote:Thanks, Q5sys, for running the tests. I can appreciate how much time they took, having recently run compartive tests on pets vs. sfses vs. program folders.

About two years ago, when PAE kernels began to show up, I did a google search and reported pretty much the same thing based on a professional lab's tests under, albeit I think, Fedora. Since then, whenever the opportunity arose, I argued that Devs should be developing two species of Puppies: non-PAE for 32-Bit, PAE for 64-bit. Regretfully, I suspect I've been preaching heresy to True Believers: neither evidence nor reason matters.Or perhaps, as I've also said, Devs are likely to acquire modern, powerful, computers and create the best system for those. The thought that a non-PAE Pup may breath new life into hundreds of thousands of computers, otherwise slated for landfills, that may be of use in the 3rd World, or to most of the people now economically depressed in the 1st --in short, that Pups actually have a social value-- no longer appears to be important.

mikesLr
No problem. I did another 4 tests this weekend focusing on the older P3 hardware. So I'll post those when I have time this week. The four tests I did were
Non-PAE idle
Non-PAE under load
PAE idle
PAE under load
The idle test was just from a simple boot and nothing else. The load test was done about 10 minutes after the idle test (same boot), with the following open and running: the web broswer, file manager, glxgears, and word processor.

If I get time I'll run the same battery on the custom i5 system this coming weekend... but I think I'm going to be rather busy so I might not be able to.

I always find running actual tests to be the way to go. Sometimes perceptions of whats fast and slow is not objective... so I prefer to stick with something that's at least a more reliable metric.

I'd be interested to read the results of your PET/SFS/FOLDER tests. Have you posted them on the forum?

User avatar
James C
Posts: 6618
Joined: Thu 26 Mar 2009, 05:12
Location: Kentucky

#4 Post by James C »

Be interesting to see your results for the older P3 era hardware. When this pae fascination began I did quite a bit of testing on the older hardware......... the results were fairly obvious from mere observation. Even a 2-3% performance hit makes a difference when a user is starting with only 256-512 mb of ram.
Looking forward to the results.

User avatar
darkcity
Posts: 2534
Joined: Sun 23 May 2010, 19:16
Location: near here
Contact:

#5 Post by darkcity »

I've just noticed this thread regarding performance penalties using PAE - so I've noted that on the Wiki.

BTW, I don't know if this has already been mentioned? But scsijon noted a developer saying PAE supports a "No Execute" feature in addition to extending memory range-

blog link at bottom of the page http://puppylinux.org/wikka/PAE

User avatar
James C
Posts: 6618
Joined: Thu 26 Mar 2009, 05:12
Location: Kentucky

#6 Post by James C »

A selection of cpu's that doesn't support PAE....
this is because the PAE mode that is required to use the NX bit causes pre-Pentium Pro (including Pentium MMX) and Celeron M and Pentium M processors without NX support to fail to boot. Other processors that do not support PAE are AMD K6 and earlier, Transmeta Crusoe, VIA C3 and earlier, and Geode GX and LX. VMware Workstation versions older than 4.0, Parallels Workstation versions older than 4.0, and Microsoft Virtual PC and Virtual Server do not support PAE on the guest
http://en.wikipedia.org/wiki/NX_bit
The support for this feature in the 64-bit mode on x86-64 CPUs was added in 2004 by Andi Kleen, and later the same year, Ingo Molnar added support for it in 32-bit mode on 64-bit CPUs. These features have been in the stable Linux kernel since release 2.6.8 in August 2004.
Why the sudden interest in something that's been around since 2004......?
Last edited by James C on Tue 26 Feb 2013, 17:32, edited 1 time in total.

User avatar
darkcity
Posts: 2534
Joined: Sun 23 May 2010, 19:16
Location: near here
Contact:

#7 Post by darkcity »

The NX bit specifically refers to bit number 63 (i.e. the most significant bit) of a 64-bit entry in the page table. If this bit is set to 0, then code can be executed from that page; if set to 1, code cannot be executed from that page, and anything residing there is assumed to be data. Also note that it is used only with Physical Address Extension (PAE) page table format, because the x86's original 32-bit page table format has no bit 63 and therefore no bit to disable/enable execution.
NX memory protection has always been available in Ubuntu for any systems that had the hardware to support it and ran the 64-bit kernel or the 32-bit server kernel. The 32-bit PAE desktop kernel (linux-image-generic-pae) in Ubuntu 9.10 and later, also provides the PAE mode needed for hardware with the NX CPU feature. For systems that lack NX hardware, the 32-bit kernels now provide an approximation of the NX CPU feature via software emulation that can help block many exploits an attacker might run from stack or heap memory.
Emphasis added. So Linux can emulate No Execute on non PAE 32bit hardware. I wonder which kernel introduced and if its typically used.

User avatar
01micko
Posts: 8741
Joined: Sat 11 Oct 2008, 13:39
Location: qld
Contact:

#8 Post by 01micko »

Q5sys,

I am not going to express my personal opinion either way on PAE/nonPAE but because Puppy is generally perceived as capable of running on older hardware I always list the 4g version first,

Now, there is a slight flaw in your tests. The 4g and PAE versions are different kernel versions in 5.4 .

For more objective testing, I suggest to use Slacko 5.3.3 because both versions ship with k3.1.10. (with 4g/PAE appended to the name).

Cheers!
Puppy Linux Blog - contact me for access

User avatar
Q5sys
Posts: 1105
Joined: Thu 11 Dec 2008, 19:49
Contact:

#9 Post by Q5sys »

01micko wrote:Q5sys,

I am not going to express my personal opinion either way on PAE/nonPAE but because Puppy is generally perceived as capable of running on older hardware I always list the 4g version first,

Now, there is a slight flaw in your tests. The 4g and PAE versions are different kernel versions in 5.4 .

For more objective testing, I suggest to use Slacko 5.3.3 because both versions ship with k3.1.10. (with 4g/PAE appended to the name).

Cheers!
I noticed the kernel version differences, but figured you had a good reason for doing so. Thats why I tested the 5.4 releases. Because even if the kernel version itself presents the performance difference... in the end the user will experience those differences between the two release versions.
So the results are real world applicable.
If I get time this weekend, (I dont know if I will). I'll try out the 5.3.3 versions and see if there's much of a difference.

Here are the additional tests I ran this weekend.
http://q5sys.info/pae/tests/newer/

User avatar
01micko
Posts: 8741
Joined: Sat 11 Oct 2008, 13:39
Location: qld
Contact:

#10 Post by 01micko »

Well if you can apply the logic "older hardware, older kernel" .. then that's my "good reason" :lol:

Which isn't necessarily true! As we all know kernel versions can vary considerably performance wise on different hardware and sometimes a newer version actually performs better than an older version on older HW, which in actual fact *should* (but isn't always) be the case.

Make sense?

Well since you are comparing 5.4 versions then I guess the real world test are relevant to 5.4 and the info is useful to anyone using or contemplating using 5.4. I may even link this thread on my 5.4.1 website, when published, so users can take a look if they want and make an informed choice.
Puppy Linux Blog - contact me for access

User avatar
mikeslr
Posts: 3890
Joined: Mon 16 Jun 2008, 21:20
Location: 500 seconds from Sol

Link to SFS files vs program folders: test results are in!

#11 Post by mikeslr »

Hi Q5sys,

I posted SFS files vs program folders: test results are in!
Here's the link http://murga-linux.com/puppy/viewtopic.php?t=84457.

mikesLr

Pelo

PAE for games only

#12 Post by Pelo »

PAE for games only. Non-PAE was running well ?
PAE is usefull when More Than 4gb of RAM needed. Which applications need that ?

gcmartin

Re: PAE for games only

#13 Post by gcmartin »

Hello @Pelo
Pelo wrote:PAE for games only. Non-PAE was running well ?
PAE is usefull when More Than 4gb of RAM needed. Which applications need that ?
Its not the applications which have requirements. PAE is NOT like 64bit where the processor is totally different. PAE is "the" processor available when an OS which understands it, uses it.

Applications make use of the services of the OS (operating system, in this case, Linux). When the kernel is built with PAE, Linux will use all RAM it finds on the motherboard because of the hardware features provided by Intel and AMD.

Thus, for example, no matter how much RAM you have it will provide it to the OS which will use what it sees for providing "space" for application execution. Developers, today, provide distro because they know that it will use all RAM, no matter if its 128MB or 64GB on the motherboard.

Because this is all done thru the hardware, their is no penalty, as it works essentially the same as the old pre1995 processors do for RAM access. In other words, 99% of the world's x86 PCs have CPUs which handle whichever 32bit distro thrown at it in exactly the same way in the hardware. One distro can use the traditional manner limited to less than 4GB, while another version can use the ALL RAM manner, provided by the processor.

Its pretty simple, but, somehow the whole topic continues to get distorted thru individual translation. There is nothing mythical or mystic. Its hardware used by the OS/distro....that's all. And, this hardware, excepting for a few Intels, has been around, working, available, built-into almost every x86 PC for about 2 decades.

Again, its offers the same RAM access as the old technology and it doesn't care how little or how much RAM you have. IT EXPOSES ALL RAM to the OS/distro.

As more and more of the community understand this, the topic has diminished in discussion. We all now know that it works! as it allows users to add as much RAM as they choose knowing that their distro will make it usefully available to the applications. Thus, they never have to change a PAE distro when they add RAM and the RAM use concern goes away.

Hope this helps in understanding

P.S. For those with some of the Intels (few "M"s used in inexpensive laptop and eeePCs) insure that YOU DO NOT ADD MORE THAN 4GB RAM. Those that have the limited 32bit CPUs cannot run a PAE distro.

Post Reply