Announcement

Collapse
No announcement yet.

Power & Memory Usage Of GNOME, KDE, LXDE & Xfce

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

  • Originally posted by mtippett View Post
    3) Present a similar set of results with the modified methodology.
    You must of read my mind.

    Comment


    • Originally posted by mugginz View Post
      You must of read my mind.
      There are some gaps. What resolution movie, screen, graphics driver, etc. The first run vs consequent run is intriguing - maybe data coversion/loading.

      But it's good to see the Analysis stage under-way (and hey, you started around post 30 or so .

      Comment


      • Originally posted by mtippett View Post
        There are some gaps. What resolution movie, screen, graphics driver, etc. The first run vs consequent run is intriguing - maybe data coversion/loading.

        But it's good to see the Analysis stage under-way (and hey, you started around post 30 or so .
        Machine was a virtual machine but will be doing runs for Celeron 1.2GHz, 512M soon.

        Complete specs as per prev post a while back.
        1024M RAM
        8G Storage
        Gfx Driver for Cirrus Logic 5446

        The movie was the demo ogg file from the Wikipedia site for Theora. I-15bis.ogg It's a 1:40 video but my script terminates playback after 20s via dbus so as to do a clean quit as per what a user might do.
        Resolution of video is 320x240 25 f/s

        Screen is set at 1024x768 so as to have 24bit/pixel (a limitation of the gfx card emulated)

        All parameters of tests will be held consistent between machines.

        Only the OS, Desktop, and hardware will change in order to see what happens for different PC configs.

        Comment


        • WHY THE MEMORY TEST RESULTS ARE WRONG

          Nobody needs to discuss the memory results as they are just wrongly calculated. The source is:

          http://www.phorogit.com/index.php?p=...fb59e352a06017

          Search for USED. It parses the output of

          Code:
          free -t -m 2>&1
          which is as example:

          Code:
          free -t -m 2>&1
                       total       used       free     shared    buffers     cached
          Mem:          1977       1231        746          0        345        359
          -/+ buffers/cache:        526       1451
          Swap:            0          0          0
          Total:        1977       1231        746
          PTS stored the value of used, but that's saidly incorrect. When you compared for example with htop the used memory is much lower. The buffers and the cache is of course "used" to speed up application lauching but will be available to apps which requrire more ram. So this has to be added.

          You get the htop value (which is correct) using:

          Code:
          free -m|awk '/Mem/{print $3-$6-$7}'
          or even without the free tool:

          Code:
          echo $((($(awk '/MemTotal/{print $2}' /proc/meminfo)-$(awk '/MemFree/{print $2}' /proc/meminfo)-$(awk '/Buffers/{print $2}' /proc/meminfo)-$(awk '/^Cached:/{print $2}' /proc/meminfo))/1024))
          Conclusion is simple: fix PTS and retest, the ram usage is just wrongly collected.

          Comment


          • Corrected source link, could not edit it...

            http://www.phorogit.com/index.php?p=...97e3a627c3e28e

            Comment


            • Originally posted by Kano View Post
              WHY THE MEMORY TEST RESULTS ARE WRONG...
              So there we have it then. It explains all. I'm still running more benches to satisfy my curiosity though.

              Comment


              • Who is the supreme ruler here ?

                Originally posted by mtippett View Post
                Completely wrong. (I was the primary author, and coined the concept - so this is one of the times that I can actually use "completely".
                OK.

                Originally posted by mtippett View Post
                The point is that the benchmarks are just what they are. Benchmarks.
                The numbers are just what they are, numbers.
                I don't dispute that. I don't dispute repeatability of those results.

                I do dispute the following: that those numbers are good representations of what is described in the Phoronix article, i.e. of the concept "memory consumption of a desktop environment". The Phoronix article was about memory consumption of DEs, but how come that those numbers entail memory consumption of for example multiple instances of Bash processes - what has Bash to do with KDE/Gnome/etc? Bash is not a part of any of those DEs.

                Or: The Phoronix numbers about the memory consumption of KDE/Gnome/etc (most likely) include memory consumption of the syslog daemon. Why? What has syslog to do with those DEs? My understanding is that "syslog" is a distinct component of the OS, disjoint and independent from the components "KDE", "Gnome", etc.

                Or: what has one-time-usage data present in the disk cache to do with using KDE/Gnome/etc?

                That is the primary problem, not whether those measurements are repeatable.

                Originally posted by mtippett View Post
                I believe that within this thread *NOONE* has gone through and reproduced *EXACTLY* what Michael did for the article. mugginz came close, and guess what, he was in a similar order.
                The issue is not whether it is repeatable.

                Originally posted by mtippett View Post
                The majority of this thread is about people arguing their side of the coin, or saying that their figures don't match up. As mentioned in the presentation, and mentioned elsewhere.
                The issue is not whether it is repeatable.

                Originally posted by mtippett View Post
                Feel free to standup and do the following
                Now, let's see...

                Originally posted by mtippett View Post
                1) Reproduce the test and the results
                I cannot. Even if I were to reproduce the test, I have a different concept of what comprises "memory consumption of a desktop environment". So my results would be different. But they sure would be repeatable.

                Originally posted by mtippett View Post
                2) Make suggestions for improvements or clarifications
                I did.

                Originally posted by mtippett View Post
                3) Present a similar set of results with the modified methodology.
                I did that already in my first post. Many people have done so as well. Seems you completely ignore that.

                Originally posted by mtippett View Post
                But I don't think I have seen anyone get past 1). No one has voting rights for 2 or 3 unless you do 1.
                Actually, many people here do have those voting rights - but you are going to completely ignore them because their results do not fit into your little nice world where everything works according to your rules.


                Originally posted by mtippett View Post
                Quotes are just that.
                ... or maybe they aren't

                Originally posted by mtippett View Post
                Out of context they can be used anyway.
                ... or maybe they should not be used like that

                Originally posted by mtippett View Post
                Fortuntately, with digging the context can be found for all the quotes.
                ... now we are all saved

                Originally posted by mtippett View Post
                Regards,

                Matthew

                Comment


                • Originally posted by mugginz View Post
                  The contents of the benchmark are as follows:
                  • Autostart on boot recording RAM usage once every second for 20 seconds.
                  • Launch Konsole and record RAM usage once every second for 20 seconds.
                  • Quit Konsole via dbus and record RAM once usage every second for 20 seconds.
                  • Launch Dragon Player with ogg theora video and record RAM usage once every second for 20 seconds.
                    (Dragon Player is the default video playback program for Kubuntu 9.10)
                  • Quit Dragon Player via dbus control and record RAM usage every second for once 20 seconds.
                  • Reboot the machine.

                  Interestingly the machine never saw more than 267M used if you dont include the first run after installation or updates (Run 1 for both Charts)
                  Interestingly, you did a similar mistake to what Phoronix did: forgetting to explicitly mention the exact method used to measure memory consumption. Since I do not know how you did the measurement, it is impossible for me to fully understand what you actually measured.

                  Ok ... Phoronix published the method by publishing the source code of PTS.

                  Once this problem is solved when (if) you disclose the measurement method, then we can start talking - about whether those numbers are accurate representations of the concept "memory consumption of a desktop environment".

                  Comment


                  • Can we please get over it?

                    KDE sucks. There, I said it.

                    Just kidding.

                    << ⚛ >>, you are beating a dead horse here, Kano found out the root of the issue and submitted a fix. The PTS tests should be rerun from scratch.

                    That said, you seem to miss the fact that syslog and similar processes add a constant amount of overhead that affects all measurements. They don't skew results one way or another when comparing DEs (as was done here).

                    Comment


                    • Originally posted by BlackStar View Post
                      << ⚛ >>, you are beating a dead horse here, Kano found out the root of the issue and submitted a fix. The PTS tests should be rerun from scratch.
                      Maybe next time the h-online will be more careful and won't blindly link to such article which gives wrong assumptions. I wonder what will be the title of the new comparison - "Power & memory usage of GNOME, KDE, LXDE & Xfce - rerun"? ;>

                      Comment


                      • Originally posted by << ⚛ >> View Post
                        Interestingly, you did a similar mistake to what Phoronix did: forgetting to explicitly mention the exact method used to measure memory consumption. Since I do not know how you did the measurement, it is impossible for me to fully understand what you actually measured.
                        That's possibly a little unfair.

                        The post I made that you reference was made at about 10 o'clock in the morning, having been up since the morning before (more than 24hrs). I got a bee in my bonnet late last night and decided to do some formal tests in order to report the results on my blog firstly, and then report them here as well.

                        I thought that before I had the full suite of numbers comparing Kubuntu 9.10 (KDE), Kubuntu 10.04 (KDE) and Ubuntu 10.04 (Gnome), I would reported some preliminary findings. I might add that the info accompanying the graphs is fairly clear though. And that it measured used memory. That is, not used memory+cache+buffers and is consistent with an earlier post I made in which I stated what I believe was the correct metric to report when stating that your benches represent used memory.

                        The used memory info is available from various places in Linux, and the one I chose was what is reported by free. To be completely clear thats the value minus buffers and cache.

                        Comment


                        • Originally posted by BlackStar View Post
                          KDE sucks. There, I said it.

                          Just kidding.
                          You trolling is infuriating, but your humor is awesome... Have to give you that!

                          Comment


                          • Indeed

                            Originally posted by BlackStar View Post
                            << ⚛ >>, you are beating a dead horse here, Kano found out the root of the issue and submitted a fix. The PTS tests should be rerun from scratch.
                            I probably found about the issue before he did. Maybe you just ignored what I wrote about htop in one of my posts.

                            Originally posted by BlackStar View Post
                            That said, you seem to miss the fact that syslog and similar processes add a constant amount of overhead that affects all measurements. They don't skew results one way or another when comparing DEs (as was done here).
                            What is your IQ, my dear? Maybe you don't know how division works. You know, the mathematical operation used in computing percentages. A citation from the Phoronix article to refresh your memory (bold typeface added):

                            Eating much more memory than the lightweight Xfce/LXDE desktops was KDE 4.4.1 that ended up leading to 67% greater memory consumption than LXDE. GNOME 2.29.1 had a 24% smaller memory footprint than KDE 4.4.1.

                            Comment


                            • Originally posted by << ⚛ >> View Post
                              What is your IQ, my dear?
                              I'm sorry, but that rhetorical question, in itself, is extremely stupid and says more about you than about BlackStar...

                              IQ is the avarage of linear measured capability of a human brain to execute tasks. So that in itself says nothing.

                              Furthermore inteligence, intelect and insight are three completely seperate things.

                              I will not for a second stop listening to BlackStar (ignoring his trolling ofcourse ) if his IQ was to be even average.

                              Every monkey, given the right education, can be a George Orwell. If someone would be extremely good at language, but has to learn a mathemathical subject for over 2 weeks and practice the questions in order to get an avarage score, one could beat a mathemathical genius in IQ scores measured today.

                              Please note that education is linked to IQ in every way...

                              True smartasses will look unbiased at anything and then form their own conclusions about that something.

                              IQ rating is for idiots that want to look smart while they are far from being so...

                              Okey next... <_<'

                              Comment


                              • Originally posted by mugginz View Post
                                That's possibly a little unfair.

                                The post I made that you reference was made at about 10 o'clock in the morning, having been up since the morning before (more than 24hrs). I got a bee in my bonnet late last night and decided to do some formal tests in order to report the results on my blog firstly, and then report them here as well.
                                OK.

                                Originally posted by mugginz View Post
                                I thought that before I had the full suite of numbers comparing Kubuntu 9.10 (KDE), Kubuntu 10.04 (KDE) and Ubuntu 10.04 (Gnome), I would reported some preliminary findings. I might add that the info accompanying the graphs is fairly clear though. And that it measured used memory. That is, not used memory+cache+buffers and is consistent with an earlier post I made in which I stated what I believe was the correct metric to report when stating that your benches represent used memory.
                                No, it is not clear to me. I agree it is maybe clear to yourself, but if you publish it subjective clarity is not sufficient.

                                "Used memory" is a very broad term. Notice there is no subject there: it is unclear who or what exactly uses that memory. Neither you, nor Phoronix, bothers to explain what is actually being measured.

                                Originally posted by mugginz View Post
                                The used memory info is available from various places in Linux, and the one I chose was what is reported by free. To be completely clear thats the value minus buffers and cache.
                                But the cache may share some physical memory pages with executables and libraries. Did you know about that?

                                So, when you do (TOTAL_USED_MEMORY - MEMORY_USED_BY_CACHE) you might in fact be subtracting away memory belonging to executables and libraries.

                                Comment

                                Working...
                                X