Can 32-bit Puppy use >3 GB of RAM? (Yes, with PAE)

[p310don]

Hi guys,

I've been having a wealth of issues with my 'net connection at home, so haven't had the chance to try out the PAE sample wary. Will get onto it as soon as I can.

Hopefully the PAE kernel won't be too significant to the performance on the old pcs.

Obviously we can see that extra gig or so in the big machines, which is where this thread is aimed, but if the performance figures are on par on the old machines, then that's something to talk to Barry about including in future "standard" puppies.

I guess the point of PAE is for those who wants to stick to 32-bit puppies because of backward of compatibility reasons (wealth of ready-made pets, etc) but still want to maximise the usage of memory.

I agree absolutely with this statement. 32bit OS's have better stability in my experience, probably due to longer development histories, but without a doubt 64bit is the future.

[James C]

Myself, I'd recommend going straight to the 64-bit path if you have more than 3GB of memory ...

For what its worth, I agree.

Since I already have a frugal install of Regular Wary 511 on my old P3 test box ( a fast 733 Mhz cpu with a whopping 256 Mb of ram and 1 Gb of swap) I just did a quick frugal install of your 511 PAE to compare.
Might be interesting......

[James C]

First interesting observation, Wary 511 PAE absolutely refuses to make a save file.Goes through the motions then doesn't save.

Copied the save file over from the regular Wary install........shouldn't matter to the kernel.

[James C]

Quick comparison before bedtime.....
Wary 511 PAE....... Firefox open here on forum.

-Computer-
Processor : Pentium III (Coppermine)
Memory : 254MB (109MB used)
Operating System : Puppy Linux 0.51
User Name : root (root)
Date/Time : Thu 07 Apr 2011 01:07:40 AM CDT

-Version-
Kernel : Linux 2.6.32.28 (i686)
Compiled : #2 SMP Mon Apr 4 20:39:31 GMT-10 2011

# free
total used free shared buffers
Mem: 254264 248564 5700 0 23308
Swap: 1020088 0 1020088
Total: 1274352 248564 1025788
#



Let me reboot into regular Wary 511......

[James C]

Regular Wary 511.... same conditions.

-Computer-
Processor : Pentium III (Coppermine)
Memory : 254MB (99MB used)
Operating System : Puppy Linux 0.51
User Name : root (root)

-Version-
Kernel : Linux 2.6.32.28 (i686)
Compiled : #1 SMP Sat Feb 5 11:15:25 GMT-8 2011


# free
total used free shared buffers
Mem: 254352 244976 9376 0 24328
Swap: 1020088 0 1020088
Total: 1274440 244976 1029464
#


Very unscientific, but it's a start.

[jamesbond]

Thank you James C. I'm not aware it refuses to create savefile (I didn't test it that far). As for memory, it's 10MB difference, quite significant for a 256MB machine (4%). How is the performance - do you feel Firefox is more/less sluggish or about equal?

p310don - take your time

[gcmartin]

@jamesbond and @James C and others, too, feel as I feel, although we may not be saying it the same way.

Year or more ago, I approached Barry, first, then the community elsewhere about the "reasonability" of declaring what platforms a PUP is targeted for.

I understand why there is a reluctance to make such a declaration. But, I feel this is a very important declaration for the reasons we are discussing this here. We are addressing a :Large RAM issue. Its important to know the amount of memory it will take on PC with small RAMs, but, we are addressing LARGE RAM PCs.

If we can demonstrate that there is no impact of these upgrades to PUPs on Large RAM PCs, then we have provided an answer to those who would/should be using this in their Distro.

And, in a separate thread, we may want to show BARRY and others how little impact this upgrade has on "small ram" PCs.

James C's report, I feel is very significant, especially to the small ram community who might find other reasons to use this distro on a small ram PC. For example, if I have small ram and this distro, I may find that even with the use of SWAP partition, my performance gain when reaching "ram saturation" (the point at which I'm using SWAP) has a much higher performance and greater stability (most important to me) than the other distro provide.

Anyway, seems that on a 512MB-8GB PC, I wouldn't concern myself with 10-20MB of storage, unless, I am using a lot of /tmp or program ram without the benefit of SWAP,

In this thread, we should put a stake in the ground on this, by "declaring" the platforms we believe this upgrade will safely deploy on Further, we should have a separate thread where those with smaller 32bit RAM platforms, that do not meet the minimum we declare, to post their findings when using this kernel. That way, issues of running this in a "small ram" on that thread, I predict, will be very different than those we discuss, here, in this "LARGE RAM" thread.

BarryK has only a small ram PC. I would think that even if this thread demonstrates reasonableness, there MAY be a reluctance on his part to want involvement.

On the other hand, other distro owners who offer solutions for easy additions to their 32bit distros (LightHouse Mariner and Official PUP V5.2 come to mind), they would want to insure that this kernel upgrade is incorporated in their distros because it is obvious that they intent this to address year 2000+ PC platforms of ALL sizes. Especaially since this really does address system stability without any significant performance execution degradation.

I think the Puppy 32bit community will benefit greatly by everyone's efforts here.

PAE hardware has been incorporated in all 32bit CPUs for many years now. It was put there for our benefit. We are now starting to see many more PC that will be adding memory in lue of upgrading. I recently saw a 4GB ECC DDR chip for $50 (US). On my 32bit Xeon, this delays a server upgrade by about a year. Others running Puppy on 32bit platform can reasonable deploy all kinds of VM and other applications totally in RAM iff their PUP distro allows it. The possibilities go on and on. Again, we benefit from this kernel distro upgrade.

Hope this helps

[p310don]

Ok, I have done my first round of benchmarks comparing Vanilla Wary 5.11 and HUGEMEM Wary 5.11

This is only very basic so far. I have booted the machines into a live cd of each version of Wary, and using Hardinfo only, have recorded the memory total and used, and then run all six benchmarks at the bottom of Hardinfo.


NON PAE (STANDARD)

CPU 2 x Athlon(tm) 64 x2 Dual Core Processor 5600+

Total Memory 3115
Ram Used 88

CPU Blowfish 7.04
CPU Cryptohash 161.29
CPU Fibonacci 2.59
CPU N-Queens 13.36
FPU-FFT 5.26
FPU-Raytracing 15.36

WITH PAE (HUGEMEM)

CPU 2 x Athlon(tm) 64 x2 Dual Core Processor 5600+

Total Memory 4147
Ram Used 87

CPU Blowfish 7.04
CPU Cryptohash 161.76
CPU Fibonacci 2.59
CPU N-Queens 13.31
FPU-FFT 5.27
FPU-Raytracing 15.34

No significant differences, apart from RAM total




NON PAE (STANDARD)

CPU CELERON (COPPERMINE)

Total Memory 189
Ram Used 77

CPU Blowfish 26.19
CPU Cryptohash 13.82
CPU Fibonacci 18.75
CPU N-Queens 44.46
FPU-FFT 88.74
FPU-Raytracing 94.34

WITH PAE (HUGEMEM)

CPU CELERON (COPPERMINE)

Total Memory 189
Ram Used 76

CPU Blowfish 26.19
CPU Cryptohash 13.43
CPU Fibonacci 18.53
CPU N-Queens 44.48
FPU-FFT 89.28
FPU-Raytracing 94.43

No significant differences.

Will do more "Real World" testing and report later. Going to the Gold Coast for a holiday now!

[gcmartin]

Again, we may want to have this thread's focus on LARGE RAM (... >3.1 GB of RAM) while starting a separate thread for PAE on "small ram" less than 512MB. That means 2 threads; this and a PAE small ram thread for discussion.

Another option is to change this thread's title to reflect the direction we are partaking. For example "PAE benefits in LARGE RAM and small ram environments" and I'm sure other tits come to mind.

Issues when using the kernel in these 2 environments MAY be different. . But it would add strength to providing Barry with reliable information which might prove useful in his evaluation of this kernel change.

This is just a hunch on my part .(I may be wrong here, though)

[James C]

Some benchmarks on a Puppy spec computer.......
Wary 511

-Computer-
Processor : Pentium III (Coppermine)
Memory : 254MB (55MB used)
Operating System : Puppy Linux 0.51
User Name : root (root)
Date/Time : Thu 07 Apr 2011 07:59:25 PM CDT
-Display-
Resolution : 1024x768 pixels
OpenGL Renderer : Mesa GLX Indirect

-CPU Blowfish
<big><b>This Machine</b></big> 731 MHz 62.186
Intel(R) Celeron(R) M processor 1.50GHz (null) 26.1876862
PowerPC 740/750 (280.00MHz) (null) 172.816713

-CPU CryptoHash
<big><b>This Machine</b></big> 731 MHz 18.923

-CPU Fibonacci-
<big><b>This Machine</b></big> 731 MHz 14.173
Intel(R) Celeron(R) M processor 1.50GHz (null) 8.1375674
PowerPC 740/750 (280.00MHz) (null) 58.07682

-CPU N-Queens-
<big><b>This Machine</b></big> 731 MHz 34.026

-FPU FFT
<big><b>This Machine</b></big> 731 MHz 67.920

-FPU Raytracing-
<big><b>This Machine</b></big> 731 MHz 72.426
Intel(R) Celeron(R) M processor 1.50GHz (null) 40.8816714
PowerPC 740/750 (280.00MHz) (null) 161.312647


Wary 511 PAE

-Computer-
Processor : Pentium III (Coppermine)
Memory : 254MB (66MB used)
Operating System : Puppy Linux 0.51
User Name : root (root)
Date/Time : Thu 07 Apr 2011 08:40:36 PM CDT
-Display-
Resolution : 1024x768 pixels
OpenGL Renderer : Mesa GLX Indirect

-CPU Blowfish-
<big><b>This Machine</b></big> 731 MHz 62.304
Intel(R) Celeron(R) M processor 1.50GHz (null) 26.1876862
PowerPC 740/750 (280.00MHz) (null) 172.816713

-CPU CryptoHash-
<big><b>This Machine</b></big> 731 MHz 18.808

-CPU Fibonacci-
<big><b>This Machine</b></big> 731 MHz 14.187
Intel(R) Celeron(R) M processor 1.50GHz (null) 8.1375674
PowerPC 740/750 (280.00MHz) (null) 58.07682

-CPU N-Queens-
<big><b>This Machine</b></big> 731 MHz 34.171

-FPU FFT-
<big><b>This Machine</b></big> 731 MHz 68.098

-FPU Raytracing-
<big><b>This Machine</b></big> 731 MHz 72.514
Intel(R) Celeron(R) M processor 1.50GHz (null) 40.8816714
PowerPC 740/750 (280.00MHz) (null) 161.312647



Unscientifically measured with a stopwatch on 3 boots each version, the PAE version takes about 7 seconds longer to boot (from selection in Grub to all icons being visible on taskbar).

The PAE Wary uses a bit more ram and is some slower on all the benchmarks.

[James C]

Year or more ago, I approached Barry, first, then the community elsewhere about the "reasonability" of declaring what platforms a PUP is targeted for.

I've been involved in the development of the Puppy "flagship" ,Lucid Puppy, since the beginning.One of the primary objectives was to keep things small enough so a user could boot the cd in a 256 mb ram computer, load everything in ram and subsequently eject the disc to free up the optical drive.
This seems to show that Puppy is still targeted toward low-spec computers but while still having the ability to run well on newer multi-core equipment.

[James C]

Thank you James C. I'm not aware it refuses to create savefile (I didn't test it that far). As for memory, it's 10MB difference, quite significant for a 256MB machine (4%). How is the performance - do you feel Firefox is more/less sluggish or about equal?

p310don - take your time

Haven't done much "real world" testing on this yet..........with only a couple of browser tabs open it seems about the same.I'm curious though what will happen with more load....I'll try to make time for some testing the next couple of days.

[p310don]

James C

Strange that you are finding more ram usage with the PAE kernel, as expected, and I am finding a tiny bit less. Do you have something else open, like firefox? I have nothing but hardinfo.

I'll have to do some boot time measurements. 7 seconds is definitely significant....

[James C]

James C

Strange that you are finding more ram usage with the PAE kernel, as expected, and I am finding a tiny bit less. Do you have something else open, like firefox? I have nothing but hardinfo.

I'll have to do some boot time measurements. 7 seconds is definitely significant....

Nope. Had nothing open but hardinfo.

Opened hardinfo, clicked summary....copied info,ran benchmark...copied info, etc.

Never tested it before but I suspect the effect on low-spec equipment is why all the major distros use a non-PAE kernel as default.

About the booting, I'm using the identical savefile (copied the regular Wary save into the PAE folder) and running an fsck on both every boot.

[jamesbond]

Thanks everyone.

I think the boot-time impact and the memory impact is quite obvious by now. Only performance impact is still dangling.

I was suggesting firefox and the like because the benchmarks from hardinfo measures mostly only the raw CPU power, which shouldn't change a lot between PAE/non-PAE.

PAE changes the way memory is managed by the OS - thus its impact will be more visible from applications that juggle memory a lot, such as seamonkey / firefox, or those that uses a lot of memory - graphics editor and large apps such as open/libre office. May be watch youtube, scroll up/down complex pages a few times (in browser/openoffice), etc.

BTW, I tested PAE Wary with shutdown - and it created a good savefile for me, hmm

[gcmartin]

Only performance impact is still dangling. ... BTW, I tested PAE Wary with shutdown - and it created a good savefile for me, hmm

I agree with JamesBond on this need. The other items are trivial (mostly non-existent on 4GB PCs).

2 Questions:
Can someone help me. Does anyone have a PAE ISO available for download that I can access for test? If so, post or send me a PM. I'm interested in testing the exact "Live" configurations you are using so that my measurements and reports can be consistent as much as possible with what's already been reported.
Edited: JamesBond's Wary. is here

Is there a hardinfo item which tells if PAE support is active in the running 32bit PUPs?.

Performance Expectations
On the performance, I want to share an experience: On a 2010 benchmark at a system center, we were not able to exercise the newer system for 2 reasons: it was so much faster that it was completing work faster than we could generate work for it? To overcome this, we had to create a timed-loop application to build enough of a load to make an attempt at comparison.
Then, on memory saturation, it was impossible due to the nature of the internet. 2 DS3 links inbound to the system failed saturate memory in such a way as to get a true measure.

Advice: We should not be too disappointed to find that the hardware PAE is just as effective at delivering pages as old frame management is. After all, the manufacturers have been doing this for over 10 years. (If fact I wouldn't be surprise if a manufacturer had only one model and doesn't tell us about it....since, in theory, a circuit design could be indistinguishable in real time.)

This community has...
JamesBond has produced a positive shockwave in the PUPs community. Thanks, enormously, for your work in helping all of us.

Everyone here has contributed as we have gone from problem identification, to measurements, to identifying the missing element(s), to producing a testbed, to testing, and hopefully (and I do mean hopefully) to 32bit ISO productions that take advantage of the finding. There are going to be many more users configuration that meet this memory model and thus, they will have a PUP solution available without thought...no matter what size they have from 512MB to 16GB.

[p310don]

Can someone help me. Does anyone have a PAE ISO available for download that I can access for test? If so, post or send me a PM. I'm interested in testing the exact "Live" configurations you are using so that my measurements and reports can be consistent as much as possible with what's already been reported.

jamesbond made hugemem wary5.11, see here http://murga-linux.com/puppy/viewtopic. ... 7&start=60

Is there a hardinfo item which tells if PAE support is active in the running 32bit PUPs?.

I didn't see it, but it may be there under the kernel section. The obvious part is if you have higher than 3100ish meg of ram, you've got the PAE kernel running.

Hope to see your results soon gcmartin. I will be posting more soon also.

[gcmartin]

Running the original 32bit WARY is immaterial based upon the work already done by this community. And, I dont have a PC that has less than 512MB of memory. So, running the non-PAE mode WARY provides no benfit to this discussion.

The resolution by JamesBond is working as expected. I see no degradation. Here's my report of running Humungous WARY o a 4GB PC!

Code: Select all

-Computer-
Processor		: 2x AMD Athlon(tm) 64 X2 Dual Core Processor 4600+
Memory		: 4150MB (159MB used)
Operating System		: Puppy Linux 0.51
User Name		: root (root)
Date/Time		: Wed 13 Apr 2011 03:17:38 AM UTC
-Display-
Resolution		: 1280x1024 pixels
OpenGL Renderer		: Mesa GLX Indirect
X11 Vendor		: The X.Org Foundation
-Multimedia-
Audio Adapter		: HDA-Intel - HDA NVidia
Audio Adapter		: USB-Audio - C-Media USB Headphone Set  
-Input Devices-
 MosArt Optical Mouse
   USB Keyboard
   USB Keyboard
 C-Media USB Headphone Set
 Power Button
 Power Button
 PC Speaker
-Printers (CUPS)-
CUPS-PDF		: <i>Default</i>
-SCSI Disks-
ATA TOSHIBA MK5055GS
TSSTcorp CDDVDW SH-S223B
Optiarc DVD RW AD-7240S
Maxtor 6 L200P0

Here's my report of running Humungous WARY on a 1.5GB laptop.

Code: Select all

-Computer-
Processor		: 2x Genuine Intel(R) CPU           T2080  @ 1.73GHz
Memory		: 1545MB (176MB used)
Operating System		: Puppy Linux 0.51
User Name		: root (root)
Date/Time		: Wed 13 Apr 2011 03:11:57 AM UTC
-Display-
Resolution		: 1440x900 pixels
OpenGL Renderer		: Mesa GLX Indirect
X11 Vendor		: The X.Org Foundation
-Multimedia-
Audio Adapter		: HDA-Intel - HDA Intel
-Input Devices-
 AT Translated Set 2 keyboard
 SynPS/2 Synaptics TouchPad
 CHICONY HP Basic USB Keyboard
 Power Button
 Sleep Button
 Lid Switch
 Power Button
 PC Speaker
-Printers (CUPS)-
CUPS-PDF		: <i>Default</i>
-SCSI Disks-
Optiarc DVD RW AD-7560A
USB 2.0 USB Flash Drive

I am unable to generate any saturation workload on iethr of these PCs.

the HardInfo benchmark tools are about the only way to measurement data.

Let me know if here is other information you would like provvided. I am not seeing any performance solwing. In fact, it seems faster.. At the very worst, its about the same.

Hope this helps

[James C]

Can someone help me. Does anyone have a PAE ISO available for download that I can access for test? If so, post or send me a PM. I'm interested in testing the exact "Live" configurations you are using so that my measurements and reports can be consistent as much as possible with what's already been reported..

Good luck on this one. As far as I know,almost all distros use a non-PAE kernel as default. The iso everyone downloads will have a non-PAE kernel, usually during an install the PAE kernel can be downloaded and installed if sufficient memory is detected. Naturally the kernel can be installed later as well, if the user adds ram for an example.
Anyway, I can't think of a distro that downloads a PAE kernel as default, except maybe Red Hat.

[gcmartin]

Good luck on this one. As far as I know,almost all distros use a non-PAE kernel as default. The iso everyone downloads will have a non-PAE kernel, usually during an install the PAE kernel can be downloaded and installed if sufficient memory is detected. Naturally the kernel can be installed later as well, if the user adds ram for an example.
Anyway, I can't think of a distro that downloads a PAE kernel as default, except maybe Red Hat.

JamesC by saying this, gives support to what we are doing.

1. We are demonstrating that running a single kernel on PC built since PentiumPro does NOT produce measurable downside in Linux operations (especially in PCs with 512MB+ RAM)
2. If Puppy distros, who understand this demonstration, adopts the PAE kernel as their standard, then there is less work of their part to have to ever address a RAM limitation as they go forward, no matter what CPU is used....as long as its PentiumPRO forward.

Puppy, truly, then becomes the RAM centric system it advertises to be no matter how much RAM in present.

Thanks for pointing that out. You are right to have helped us with that understanding.

Hope this helps.
P.S. In my eyes, this is historical in the Puppy world. Almost like discovering the "world is not flat"