Announcement

Collapse
No announcement yet.

Would Phoronix do Gentoo Linux versus Ubuntu Linux benchmarks?

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • #31
    Originally posted by Shining Arcanine View Post
    I forgot to mention, I saw a link to that webpage on the gentoo forums two months ago and read it in detail. While I found it to be hilarious, it is only a parody of things and not a perfect reflection of reality. If I recall correctly, the jokes on that webpage revolve around "-O4 -fultimate-optimization -fopt-a -fopt-b -fopt-c..." and many other pointless, redundant or non-existant optimizations such as those.
    Many of the quotes are there because they display Gentoo triumphalism founded on extremely shaky ground. Stuff like "binary packages are almost useless"; "no RPM-based distro can do that [offer different compile-time options]"; "when it comes to dependencies [on Debian], you probably are still going to have to get them all on your own" (though that last one seems likely to be trolling).

    Setting your system CFLAGS is less than 0.1% of the stuff you will do to get a working system and while it is important that your system CFLAGS are configured correctly, Gentoo attempts not to maximize performance of your system, but to maximize the customizability of your system. Experiencing higher performance is often a side effect of that, where the choice of CFLAGS being one of the customization options you have when using Gentoo.
    That is why I think it's meaningless to "benchmark Gentoo": my Gentoo is not your Gentoo is not Michael's Gentoo. In principle we could all agree on a configuration to test, but nobody would actually use it, so who cares how it benchmarks?

    Comment


    • #32
      I had been wondering some time ago if it would worth installing (creating) a gentoo onto my machine. Since I do engineering calculations some speedup would be really nice. But eventually I came to the conclusion that even if my machine would be faster I would still loose more CPU cycles on compiling stuff.

      I mean if my machine is 10% faster but spends a day per every week compiling stuff then what's the point?
      I'm interested in what actual users think about this.

      Please bear in my that this is just my "feeling" about the topic mostly based on what I've read.

      Comment


      • #33
        I usually update my desktop and other machines once a week. My desktop is unstable, the other three machines are stable.

        On my experience in seven days on the unstable you get about 20-40 packages to update. With Quadcore this is about 15min of compile time.

        On the stable maybe 1-10. On the smallest machine 1.7 GHz Celeron it really depends. It got a mysql update 2 days ago, took it about 30min to build.

        This is not set in stone, it depends on which package gets updated and when. I don't see that big of a problem with 15 to 30 minutes compiling a week.

        Comment


        • #34
          //edit: anyway, you can still use the programs! The system compiles the binaries in the background and you can continue to work, once the binary is finished to build in the "sandbox" it will be moved to the real system and replaces the existing ones. Then you e.g. restart mysql and that's it...

          Comment


          • #35
            Originally posted by HokTar View Post
            I had been wondering some time ago if it would worth installing (creating) a gentoo onto my machine. Since I do engineering calculations some speedup would be really nice. But eventually I came to the conclusion that even if my machine would be faster I would still loose more CPU cycles on compiling stuff.

            I wouldn't bother. If you have the option to obtain the source of the packages you use, compile them. Different compilers will also affect performance, try different ones if you have access to them. But there's no need to compile the whole system from source. As an example, one of the most widely used quantum-mechanical calculation packages is provided both in binary and source format, so you can tweak the compilation to your needs. The supported Linux distributions are Suse and RedHat. This, of course, doesn't mean that it wouldn't run in others, although at this level you probably want to make sure you have what is more likely to work. At any rate, the people I know using this sort of thing don't even think about fucking around with Gentoo. They actually have work to do--them and their machines.

            Comment


            • #36
              Originally posted by disi View Post
              //edit: anyway, you can still use the programs! The system compiles the binaries in the background and you can continue to work, once the binary is finished to build in the "sandbox" it will be moved to the real system and replaces the existing ones. Then you e.g. restart mysql and that's it...
              And if your keyboard locks up when compiling something, check you're using evdev and not kbd as the xorg keyboard driver.
              Just wanted to mention that seeing as I finally found and fixed that issue a few days back!

              Comment


              • #37
                Originally posted by Shining Arcanine View Post
                In a fair comparison (all important software versions being the same), Gentoo likely would win in benchmarks. It does not derive its name from the fastest species of penguin in the world for nothing.
                Originally posted by Shining Arcanine View Post
                [...]Gentoo attempts not to maximize performance of your system, but to maximize the customizability of your system.
                So which one is it? Make up your mind.

                As I said in a post above, don't go into Gentoo thinking it will get you better performance.

                Where did you obtain your information regarding what Gentoo users ridicule?
                I was talking about funroll-loops ridiculing the ricer mentality of misguided Gentoo users:
                This page is dedicated to the Linux Community's greatest ambassadors, Gentoo users. Like the annoying teenager next door with a 90hp import sporting a 6 foot tall bolt-on wing, Gentoo users are proof that society is best served by roving gangs of armed vigilantes, dishing out swift, cold justice with baseball bats to those fucking ricer bastards.
                Yes, this is the very mentality that leads to "Gentoo represents the upper limit of performance" fallacies.

                Comment


                • #38
                  Fraking 1 minute edit window

                  Originally posted by Shining Arcanine View Post
                  Originally posted by BlackStar
                  Did you even read the linked article? There were tests where using the "better" -march resulted in a 50% performance degradation.

                  The funroll-loops quotation certainly fits, as you seem to be going into Gentoo with the very attitude they ridicule (Gentoo = CFLAGS = moar speed!)
                  I am actually a regular on the forums and the general consensus is to use CFLAGS="-O2 -march=native -pipe" on amd64 platforms and possibly appending "-fomit-frame-pointer" on x86 platforms.
                  So I guess you did read this article. If so, I'd suggest taking a long, hard look at the very first test (Ogg Encoding), where this "-march=native" suggestion made the code run 50% slower.

                  There's much, much more to optimization than playing with CFLAGS. Gentoo won't magically run faster than a different distro because you used a specific flag, real-life performance is much more complicated than that.

                  Comment


                  • #39
                    Originally posted by BlackStar View Post
                    So I guess you did read this article. If so, I'd suggest taking a long, hard look at the very first test (Ogg Encoding), where this "-march=native" suggestion made the code run 50% slower.

                    There's much, much more to optimization than playing with CFLAGS. Gentoo won't magically run faster than a different distro because you used a specific flag, real-life performance is much more complicated than that.
                    I did read that article. The consensus at the Gentoo forums was that the benchmark results were flawed:

                    http://forums.gentoo.org/viewtopic-t-817950.html

                    I think I referenced that in my first post in this thread.

                    Comment


                    • #40
                      Originally posted by HokTar View Post
                      I had been wondering some time ago if it would worth installing (creating) a gentoo onto my machine. Since I do engineering calculations some speedup would be really nice. But eventually I came to the conclusion that even if my machine would be faster I would still loose more CPU cycles on compiling stuff.

                      I mean if my machine is 10% faster but spends a day per every week compiling stuff then what's the point?
                      I'm interested in what actual users think about this.

                      Please bear in my that this is just my "feeling" about the topic mostly based on what I've read.
                      Your software (that I assume you compiled yourself) will likely not be any faster simply because you are running Gentoo. The only benefits Gentoo might have (which you can get with your current distribution) are an optimized kernel (optimizing it specifically for your application would be a good idea), which you can compile on your current system and fewer background services, which you can get by disabling stuff on your current system. You might see some improvement from stuff not needing to be compiled into background services that do need to be run in the background, but that improvement would mostly be restricted to start-up time and memory usage.

                      If you run a desktop environment, you would probably be better off by not running it when you use your application, but if you insist on running it (in which case you would have to decide whether your application or your desktop environment must suffer from kernel optimizations), some of the bloat that it involves could probably be removed by using Gentoo, because it lets you strip out stuff you don't want/use, which can be helpful in the case of features that siphon CPU cycles from yoru system, like strigi and nepomuk.

                      In short, your software will not become magically faster because of Gentoo. It will only benefit indirectly from things (i.e. customization) that Gentoo encourages, which you should have done all along.

                      Comment


                      • #41
                        Thank you guys for the feedback.

                        Shining Arcanine: I have to admit that the softwares I run are proprietary ones. You know that writing FEA and CFD codes are rather complicated. (Though I have to mention that the capabilities of OpenFOAM are remarkable.)
                        I even feel fortunate because there are linux versions for these (e.g.: ANSYS) unlike CAD applications (there are no more PRO/E or CATIA support for *nix).

                        Point is that I want to use my system without bothering much with upgrades, dependencies, etc. So Ubuntu is good for me because generally it is stable, gives me more freedom than a certain other OS and most importantly causes less pain.

                        The problem for/with the above mentioned programs is the lack of list of dependencies. Even if they say that Ubuntu is supported I often end up manually installing a few libraries. There should be a standard set of libraries which are included in every flavour of distros.

                        Comment


                        • #42
                          Originally posted by Shining Arcanine View Post
                          I did read that article. The consensus at the Gentoo forums was that the benchmark results were flawed:

                          http://forums.gentoo.org/viewtopic-t-817950.html
                          Yeah, all four posters conceded that the benchmark was flawed.

                          Comment


                          • #43
                            So, I just tested a few things
                            While messing around I had a strange thing. Here are results of i7 Ubuntu vs phenom Gentoo and all the compilation is not displayed for my phenom:

                            http://global.phoronix-test-suite.co...72-18029-25142

                            It definately did something, but came up with average time 0 seconds all the time... what did I do wrong?

                            Here is a compilebench test, which worked fine:
                            http://global.phoronix-test-suite.co...015-26573-4799

                            Comment


                            • #44
                              Originally posted by HokTar View Post
                              Thank you guys for the feedback.

                              Shining Arcanine: I have to admit that the softwares I run are proprietary ones. You know that writing FEA and CFD codes are rather complicated. (Though I have to mention that the capabilities of OpenFOAM are remarkable.)
                              I even feel fortunate because there are linux versions for these (e.g.: ANSYS) unlike CAD applications (there are no more PRO/E or CATIA support for *nix).

                              Point is that I want to use my system without bothering much with upgrades, dependencies, etc. So Ubuntu is good for me because generally it is stable, gives me more freedom than a certain other OS and most importantly causes less pain.

                              The problem for/with the above mentioned programs is the lack of list of dependencies. Even if they say that Ubuntu is supported I often end up manually installing a few libraries. There should be a standard set of libraries which are included in every flavour of distros.
                              Gentoo does dependency resolution for you, so you really do not have to bother with it. The software you manually install is probably available on Gentoo. You will just need to tell Gentoo to install it with "emerge <package-name-here>" and it will automate fetching the package, compiling it and installing it.

                              The only behind the scenes thing with which you will need to concern yourself is Gentoo's masking system. Gentoo has three levels of packages, stable, testing (softmasked) and masked (hardmasked). Stable software is what is installed by default and Gentoo always installs the latest version of this unless you tell it to do otherwise. Testing software is considered stable by upstream, but it has not yet been cleared by Gentoo's package maintainers as being stable. Most Gentoo users instruct Gentoo to use this software by default without issues, myself included.

                              If you find that some of the packages Gentoo provides are out of date and you need to start unsoftmasking stuff, dependency resolution tends to break, which requires that you unmask successively greater numbers of packages to get things to work. This never damages your system because the package manager will not go ahead with upgrades/software installations unless all dependencies have been fulfilled, but this situation can be avoided by running as ~arch as Gentoo users call it. This requires that you add ACCEPT_KEYWORDS="~x86" to your /etc/make.conf file, assuming x86 is your architecture (~amd64 is used if you run x86_64) and install the latest updates for your system. Since this will upgrade some low level stuff, you will need to follow Gentoo's OpenRC migration guide if you do this to make sure everything works correctly. After doing this, dependency resolution should always be done correctly by Gentoo's package manager and you will not have to concern yourself with whether or not some dependency is in the stable tree or the testing tree.

                              Lastly, the masked packages are packages that either are known to have something wrong with them or are beta releases from the upstream developers. These you usually never need, but some Gentoo users use them. In the case of my laptop, I am using the latest masked versions of the nvidia-drivers, chromium and opera packages, which are masked because they are considered betas by upstream, but all of them are known for being release quality software, despite being considered beta software by upstream. I have my own reasons for needing things to be this bleeding edge in these 3 cases.

                              Once you explicitly install a package via Gentoo's package manager (as opposed to an implicit installation where something is installed because something you specified either directly or indirectly depends on it), it is recorded in Gentoo's /var/lib/portage/world file, so any time you have Gentoo check for updates for your system, Gentoo will check for updates for these packages for you, which I find makes things very convenient. If you do not want certain software to be updated beyond a specific version, you can tell Gentoo that in /etc/portage/package.mask by specifying either the package and its version or a range of versions (using >= or <= as a prefix) and Gentoo's package manager will obey that.

                              Comment


                              • #45
                                I would love to see a rolling distro shootout Gentoo vs. Arch vs. Debian Sid But somehow I doubt a benchmark like that will ever be done on Phoronix...

                                Comment

                                Working...
                                X