Testing the speed of linux distributions

[vvx]

Well, I've decided that I've heard enough about Gentoo, it's time to try it. I like the idea of software compiled with the options I use. To a lesser extent, I can also see a speed advantage with software compiled for your hardware specifically. So, I'm going to put gentoo on my laptop as well as RH9 and I'd like to do some speed tests. I haven't really given it much thought yet, but I've decided it will be an automated script (I don't trust myself to run the commands at the right time etc to do a fair comparison). Some of the things I was thinking about were creating/removing directories, touching files (okay, this is real basic), resizing and doing affects on imagemagick, perhaps encoding a short divx or xvid file, using gimp scripting, etc..

I plan on using only redhat packages on the redhat machine and whatever I can emerge on the gentoo machine, so some of those might not be doable (don't know if there's anything to encode divx on a redhat package.. though, maybe I could just encode oggs instead.)

So, anyone have good ideas for things to run and compare speed between the two?

[Tyr_7BE]

When setting up the gentoo system, make sure you read up on optimizations! Remember that everything in Gentoo has to be hand-tuned. It's like a hand-crafted lambourghini...screaming fast and customized to hell, but it's a labour of love. Check out options to put in your make.conf file (like compiling with -O3 and -funroll_loops and other optimizations), and put in every USE flag that you can to make things fair (since I believe redhat comes with the works). If you don't know what these mean, there's lots of documentation on the gentoo site. It's one of the most well documented distros I've seen.

You'll also manually have to start hdparm. There's lots of documentation available for it and I'm pretty sure it's in gentoo's portage system. If you're not familiar with it, it's a program that dramatically improves memory access time in linux by using things like DMA. I know RedHat installs it with some default settings, so you might want to check out what those are with "hdparm /dev/hdX" and duplicate them on gentoo. Just trying to level the balance here

For speed it's difficult to say. You have to be certain to try a wide range because RedHat tampers with a lot of its software. For example, I believe the version of gcc they include with their distro has redhat-specific extensions, as does their X server, and I know they've done a lot of work on their OpenOffice.org.

Anyway man let us know the results!

[ifred]

You can also check out this link for more ideas:
http://articles.linmagau.org/modules...227&page=1

[vvx]

[quote author=Tyr_7BE link=board=7;threadid=7659;start=0#msg69868 date=1062208943]
When setting up the gentoo system, make sure you read up on optimizations! Remember that everything in Gentoo has to be hand-tuned. It's like a hand-crafted lambourghini...screaming fast and customized to hell, but it's a labour of love. Check out options to put in your make.conf file (like compiling with -O3 and -funroll_loops and other optimizations), and put in every USE flag that you can to make things fair (since I believe redhat comes with the works). If you don't know what these mean, there's lots of documentation on the gentoo site. It's one of the most well documented distros I've seen.

You'll also manually have to start hdparm. There's lots of documentation available for it and I'm pretty sure it's in gentoo's portage system. If you're not familiar with it, it's a program that dramatically improves memory access time in linux by using things like DMA. I know RedHat installs it with some default settings, so you might want to check out what those are with "hdparm /dev/hdX" and duplicate them on gentoo. Just trying to level the balance here

For speed it's difficult to say. You have to be certain to try a wide range because RedHat tampers with a lot of its software. For example, I believe the version of gcc they include with their distro has redhat-specific extensions, as does their X server, and I know they've done a lot of work on their OpenOffice.org.

Anyway man let us know the results!
[/quote]

Well, it seems using every USE flag would give RH the advantage, no? Surely a piece of software would be a bit more optimized if you only compiled support for gnome rather than gnome and KDE (okay, crude example) I figure I will set both up the way I would use them. Defintly will use the optimizations that I can on Gentoo, as that is the whole point here.. As for RH's tweaks, it would be interesting to see how they fair against custom compiled stuff on gentoo, which is the whole point of course.

[vvx]

[quote author=ifred link=board=7;threadid=7659;start=0#msg69870 date=1062212440]
You can also check out this link for more ideas:
http://articles.linmagau.org/modules...227&page=1
[/quote]

Some good ideas here, and some that I think would be hard to script.. Like the one test where they're talking about 3.8s vs 4.1s .. I'm not that quick with a stop watch! Far easier for me to run date a few times in the tests to get automated times without stopwatch (though, only to nearest second)

[sarah31]

may i suggest that if you are going to compare sped that you at least optimize your gentoo with common cflags (ie find out what other distros use) and then go from there.

i think often we forget that the reasons optimizations are "lacking" in precompiled binary distros is that they design their systems to be easily usable by people of a vareity of skill and on a variety of computers. more optimized and source distros compromise user numbers for "flexibility".

comparing i386 to i686 and other such optimizations is kinda silly in the end it is like comparing a k7 to a k6. of course one is better in almost aspects.

so i guess i am pleading that you compare apple to apple and not aples to oranges as much as possible.

[vvx]

If I compile it with similar optimizations (not very optimized) to other distributions in order to have an "apples versus apples" comparison, I'll end up showing that software compiled with similar options is roughly equal.. Does anyone really use gentoo with the intent of that? I don't think so.

So, I will be comparing apples to oranges. At least, if you look at the fact I'm compiling software with more optimizations on gentoo than the stock binary distributions tend to do.

If you instead consider the fact I'm comparing redhat in a typical redhat setup to gentoo in a typical gentoo setup it's apples to apples though.

Really, comparing gentoo in a handicapped setup would be self-defeating. Show that a system in a configuration that nobody would use is or isn't faster? What's the point, nobody uses that configuration!

BTW, I'm rooting for redhat!

[vvx]

I might as well add tonight's comments in while I'm here..

gentoo kicked my ass in round 1.

I downloaded the livecd off a nice fast mirror, read the documentation, etc.. When it got time to install, it was all going great except I couldn't get my orinoco wireless card to work. Nothing in the docs about problems with wireless cards, so I figured it would just work.

Searched through gentoo's forum and found that the livecd is missing a module required for the orinoco cards (or perhaps an option in the kernel I don't remember.) The suggested workaround was to use knoppix to install gentoo.

But, not all is lost -- you can do a stage3 build using just the contents of the livecd.

Except, the tarball wouldn't untar -- gave me errors. The iso passed the md5 before I burned and the read test nero does after I burned so I don't know what's up with that. Oh well.

Now downloading knoppix so I can install gentoo.

If I was writing a review on the ease of install gentoo would be getting docked some serious points!

[Tyr_7BE]

Yeah...the gentoo install is legendary.

[sarah31]

well oranges to apple tells me nothing apples to apples tells me that at the least they have set they system up in such a way as to produce maximum efficiency.

for your information i ran an -march=k7-athlon (or something like that whatever my processor was at the time) -O2 -fomit-frampointer and all that was faster for me was the performance of my protien cruncher. Libranet noticable out performed it in every other aspect.

there is an apple to oranges for you.

alot of time speed does not depend on optimizations.

to me speed counts for little.