Terrible Linux performance, great Win32 [SOLVED? - see post]

[Vla]

You definetly should activate Highmem support in your kernel or use one with 1 Gig low-mem support.
Maybe you want to poste your kernel conf?
Tried adding APIC to your config? Might be usefull to avoid all this IRQ11 sharing.

[saber850]

Here's one of my machines for reference:

[Felixlein]

Hmmmm,strange strange strange;made a few tests with my desktop-box and my laptop,

here the lappi(prescott,1Gb RAM,standard clock):

samuelcolt felixlein # bashmark
#######################################################
: T E S T : :S C O R E : : R A T I O:
:-----------------------------------------------------:
:Cpu, Integer : : 914: : -15%:
:Cpu, Floating point : : 21: : -97%:
: : : : : :
:Memory r/w (cached) : : 645: : -46%:
:Memory de-/alloc : : 233: : -64%:
: : : : : :
:Multithreading : : 686: : -72%:
#######################################################
: S Y S T E M I N F O :
-------------------------------------------------------
2x Intel(R) Pentium(R) 4 3065.563MHz, L2 512KB
Linux 2.6.15-gentoo
GCC 4.1.0-beta20060106
88KB binary size
######################################################

(no throttling or scaling or speedstepping at all,sure as hell!)


Here the desktop (clawhammer,1Gb Ram, standard clock):

automat ~ # /etc/init.d/cpudyn stop
* Stopping cpudynd ... [ ok ]

automat ~ # bashmark
#######################################################
: T E S T : :S C O R E : : R A T I O:
:-----------------------------------------------------:
:Cpu, Integer : : 1571: : +46%:
:Cpu, Floating point : : 1461: : +92%:
: : : : : :
:Memory r/w (cached) : : 12023: : +899%:
:Memory de-/alloc : : 779: : +19%:
: : : : : :
:Multithreading : : 2562: : +3%:
#######################################################
: S Y S T E M I N F O :
-------------------------------------------------------
1x AMD Athlon(tm) 64 3400+ 2210.784MHz, L2 1024KB
Linux 2.6.14-gentoo-r4
GCC 3.4.4
108KB binary size
#######################################################
: R E F E R E N C E S Y S T E M I N F O :
-------------------------------------------------------
Reference system was Geno's pc with:
Athlon XP 1800+ 1575.631MHz, 256KB
Linux 2.6.11-ck1
GCC 3.4.3-20050110 (compiled with standard cflags)
glibc 2.3.4 (with nptl)
128KB binary size



But real-life-performance of this two machines is almost equal!!!
did some serious untarring,emerging and movieencoding with this two
and none really sucked,the AMD is just a little faster,but more or less
marginal...

USE-flags are exactly the same on both machines.

maybe a gcc-issue (at least in my case)???
will check on this (have a system-backup of my laptop with gcc 3.4.4)
but no time for this at the moment.

[Suer7reus]

Thank you all so much for your replies, and please forgive me for getting back to you so late! I have unfortunately had some unrelated connectivity issues.
Anyway, my conclusion is that my problem is largely unrelated to my bashmark performance because I have decided that bashmark is a poor test of system speed (for whatever reason). People's results (including my own in win32) make so little sense I can't help but conclude that there is not absolute relationship between poor bashmark performance and poor system performance. I now intend to redo all my prior troubleshooting and test my steps with lots of actual system use instead of mainly bashmark. My minimal testing kernel, for instance, was not capable of starting X due to a complete lack of driver support. As for the kernel options you suggested I enable (Highmen, APIC, etc) they are enabled in my normal configuration, and I will ensure that they remain so as I reconfigure things for testing purposes. I will also certainly take the suggested IRQ command line options into consideration. Thank you all for your help! I think this thread is closed now, given how much trougleshooting remains to be done (in light of the ineffectiveness of my previous testing). I will post when/if I resolve my performance issue.

[philc909]

I'm having the exact same problem. The problems appeared after I emerged freenx and xorg-x11 so I initially posted the problem under Portage & Programming:
http://forums.gentoo.org/viewtopic-t-423606.html

My laptop (problem system) pentium 4 at 2.4GHz gives me the following bashmark and as bad nbench scores (see other forum)

[revertex]

[p.n]

Suer7reus have you tried nbench for benchmarking? Does it give you the same inconclusive results?
_________________
"Every time I learn something new it pushes some old stuff out of my brain." - Homer Simpson

[saber850]

I find Open Benchmark Lab's CPU benchmark pretty good:
http://www.openbench.com/benchmark.html (scroll down for the download).

[Suer7reus]

I'm working on trying (and comparing) these two benchmarks as I type - thanks for your replies! In case the USE crowd is still interested, I've converted my flags from make.conf to packages.use. Now make.conf just has the use.desc flags I've selected and a few use.local.desc's that flagged several packages. All hail grep.

[HecHacker1]

i'm curious.. what filesystems do you have? I had major breakage with reiser4 around that 2.6.12 nitro kernel. Everything worked fine, but it was super slow and disk access was horrible. I switched to ext3 until resier4 is cleaned up.

[Suer7reus]

Well, actually my problem has now been solved for good by Dell, who after four motherboard replacements finally decided it would be a good time to send me a replacement computer, an M170, with slightly better specs and much, much, much, MUCH better real performance. It is shiny, and I love it.

As for filesystems, I have for a while now used reiserfs (3) on my root partition, NTFS for Windows, and shared an ext3 (mounted as ext2 in Windows with drivers available at http://ext2fsd.sf.net) partition between them loaded with home folders and games, creatively symlinked and mount --bind'ed. That worked very well, especially for sharing settings between Windows and Linux apps (like Firefox, etc), but unless I need Windows pretty badly before the end of the week I am probably going to get rid of it! I would like to try reiser4, but the new laptop has an Intel wireless card which is kind of a pain to deal with using nitro kernels, so I have since switched back to gentoo-sources and will wait on the reiser4 for a while at least.

[saber850]

My system (P3 1GHz; 1GB RDRAM) got a noticable speed boost by recompiling everything with GCC 4. I don't recommend this unless you have some programming knowledge/experience since some packages don't build w/ GCC 4 yet (it's more strict than v 3.x).

I've narrowed down most of the performance bottlenecks to in X. I suspect X is not expoiting the hardware acceleration of my video card (GeForce FX 5200; 128MB RAM) very well. I recently found some documentation that suggests using nVidia's AGP driver as opposed to the kernel's (AGPGART) for certain chipsets (including mine: Intel 820). I've reconfigured my system to try this and am awaiting a convenient time to reboot.

Regarding filesystems, I'm currently using ReiserFS 3 on 2 machines and ext3 on another. I've come across posts that say ReiserFS 3 fragments and causes poorer performance (than ext3). I plan to try JFS or XFS next.

[Suer7reus]

I am using GCC 4 on my new system - it works great for almost everything and gcc-config'ing back to 3.4.5 to fix (and it does) the occasional problem is easy. My old system wouldn't ever build any GCC 4.x version, but retrospectively it was sort of a piece of shit.

I agree that X seems to be a big performance hog, having only really looked around with top and my Gnome system monitor, but I have on this system switched completely to Xgl (which in combination with my system's built-in LED's, actually - no joke - picks up chicks), with a secondary unmanaged Xorg server started by gdm for games and things that don't always play nicely with Xgl. Xgl, in addition to being one of the sexiest things I can think of, runs very fast and very smoothly without even denting my performance numbers. I can't advise you regarding AGP as nVidia card is PCIE =). Good luck though!

I wish I could play with filesystems more than I do - my system configuration and partition layout are awfully resilient to trashing and playing with partitions, but until I go no-Win32 I won't really have much leeway due to compatibility and space concerns. Plus filesystem changes take forever what with all the copying and moving of data from place to place, and until classes end at the end of the month, time is of the essence.
I've used Reiser 3 reflexively on a number of desktops, notebooks, and a couple of servers, and as far as I can tell it's worked great, but I don't think I'd notice anything wrong unless it was pretty drastic.

[troinfo]

Ok,

I tried this bashmark thing and here are my results:

[Suer7reus]

No; we pretty much decided that bashmark is a very subjective test. As long as your system seems to run well, your numbers are not unusually low relative to others posted in this thread - despite Geno's magical super-PC's indication otherwise. Might be a good time to upgrade GCC though =). (You may have to use gcc-config to select a newer compiler profile like 3.4.6). Here are my new numbers, but bear in mind this is an entirely different system than the one this post covered:

#######################################################
: T E S T : :S C O R E : : R A T I O:
:-----------------------------------------------------:
:Cpu, Integer : : 1693: : +58%:
:Cpu, Floating point : : 71: : -91%:
: : : : : :
:Memory r/w (cached) : : 949: : -21%:
:Memory de-/alloc : : 680: : +4%:
: : : : : :
:Multithreading : : 1910: : -23%:
#######################################################
: S Y S T E M I N F O :
-------------------------------------------------------
1x Intel(R) Pentium(R) M 2133.000MHz, L2 2048KB
Linux 2.6.17-rc4-tycho-r1
GCC 4.1.1
72KB binary size
#######################################################
: R E F E R E N C E S Y S T E M I N F O :
-------------------------------------------------------
Reference system was Geno's pc with:
Athlon XP 1800+ 1575.631MHz, 256KB
Linux 2.6.11-ck1
GCC 3.4.3-20050110 (compiled with standard cflags)
glibc 2.3.4 (with nptl)
128KB binary size
Scores gathered on March, 30th. 2005 with bashmark 0.6

This system replaced the other, faulty one. It flies.

[kartebi]

here is my mark

[Gentree]

a few thoughts on filesystems.

be aware the reiserfs defrags badly with use and slows down a lot. No defrag so you need to cp or tar the whole lot elsewhere to clean it up. Neither is it that fast to start with. Even namesys dont develop/support it any more. I dont see point in using it now.


R4. I used nitro for a couple of years on various nitro kernels (on athlonXP). I now use "no-sources" but if you have probs with compat and just want R4, patch your gentoo-sources. It's a doddle.

I just did this on an elive installation (ubuntu-deb derivative) , got the patchs direct from namesys , untar, patch, make menuconfig and turned on the R4 module. I could not believe how well it went and how easy it was.

Slightly faster then ext3 , the main thing I like is atomic writes. Very resistant to power outs.


ext3 for Win32 . Dont.

I regard this as very ill-advised unless you want to open a vector for windows crap to mess with your linux installation.

It is fine if you want to use ext3 on a windows only box and that's all.

The only way I would accept that on dual-boot is if you dont use ANY ext2/3 on linux.

I use vfat for any shared data partitions.

While on the subject , check if you have misc-binaires off in the kernel , I recently saw a post on Ubuntu (which has this on by default!) were a user had "acidentaly" run googleearth as root and it had spawned IE6... as root!

You are clearly a few orders of magnitude clever than that guy but you get the point about unnecessarily opening up vectors to bad/malicious wincrap.

_________________
Linux, because I'd rather own a free OS than steal one that's not worth paying for.

KX7-333 , AthlonXP1800+ @2.3GHz
AthlonXP-M on A7N8X @ 2.6/2.4GHz (winter/summer)
2.6.32-hh1 : portage ~x86

[suicidal_orange_II]

Having read through this whole (thankfully short) thread just incase there was any performance gain to be had I decided to see how my pentium d system compared. My bashmark came out at

[optiluca]

[curmudgeon]

[ripperd]

For those that care, an AMD64 natively compiled and ran system:

With an idle system: