Announcement

Collapse
No announcement yet.

The Huge Disaster Within The Linux 2.6.35 Kernel

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

  • Originally posted by cynyr View Post
    so any sort of hint as to what was the problem or a link to a LKML discussion of it?
    http://marc.info/?l=linux-kernel&m=127534862110443&w=2

    http://marc.info/?l=linux-kernel&m=127478957817653&w=2

    http://marc.info/?l=linux-kernel&m=127502533400323&w=2

    http://marc.info/?l=linux-kernel&m=127455736017861&w=2

    Comment


    • So this was udev consuming 100% cpu? If it was, no one looked at top and iotop before running the benchmarks, to make sure the system was truly idle before running them?

      Earlier it was mentioned that it may not have been anno_inode + udev causing the problem, but no one confirmed that top and iotop looked okay on the systems running the benchmarks. It would have been the first thing i looked at with a massive drop like that, "did one of the tests get out of control and stay alive, consuming cpu?"

      Also much hate towards the Motorola popover that covers the entry box.

      Comment


      • Originally posted by Michael View Post
        To those that are saying it's due to a udev issue, that is not for certain yet. Linus is looking into this regression now.
        Ahh found the quote i was looking for. Add the above to my above post since it took me to long to find it to edit my post.

        Comment


        • @cynyr,
          Briliant! Now if only you have figured it out before it was figured out.

          Something about credit where it's due and people eagerly want to climb up some kind of a social acceptance ladder and primal instinctive behavior...

          Comment


          • Revised with direct questions.

            Originally posted by V!NCENT View Post
            @cynyr,
            Briliant! Now if only you have figured it out before it was figured out.

            Something about credit where it's due and people eagerly want to climb up some kind of a social acceptance ladder and primal instinctive behavior...
            Credit is due to whom ever alerted kernel devs via normal methods(LKML) first.

            I'm guessing my previous post was not taken the way I meant it. I'll revise it below, in simple yes/no questions:
            1. Was the anon_inode bug linked most recently by energyman, the problem at the start of the article?(Michael mentioned that he did not know for sure.)
            2. Does PTS check cpu and IO loads before beginning a test?
              • If it does, does PTS tag the results with that information?
              • If PTS does not check, could it be made to?
            3. Why wasn't idle CPU/IO load looked at/into prior to posting this article?
              • If it was, why was it left out of the article?
              • Is not ensuring an idle system before running tests SOP for Phoronix?
            I'm not saying that PTS is not a useful tool, simply that like all tools they need to be used with cation and applied appropriately.

            The bit about the fullpage popup(flyup?) that is getting though adblock still stands unrevised. It is very hard to enter text in the box when it is covered by a popup.

            Comment


            • @Cynyr,
              What triggered my post was "the first thing I would have looked at" part. As if you implied (my words): "I would have done it ten times better", to which I was like "but you didn't".

              Comment


              • Originally posted by V!NCENT View Post
                @Cynyr,
                What triggered my post was "the first thing I would have looked at" part. As if you implied (my words): "I would have done it ten times better", to which I was like "but you didn't".
                ahh, i guess my point was, i look at top (not usually iotop) before running even a simple benchmark like hdparm, if i care about having decent results. I would expect a test suite designed for benchmarking a large number of systems in to check for an idle system prior to running tests.

                The thing that rubbed me about this, is that if SOP here isn't to look at those sorts of things to ensure a consistent test setup prior to running benchmarks, then all benchmark results are going to be suspect IMHO.

                Comment


                • Originally posted by cynyr View Post
                  ahh, i guess my point was, i look at top (not usually iotop) before running even a simple benchmark like hdparm, if i care about having decent results. I would expect a test suite designed for benchmarking a large number of systems in to check for an idle system prior to running tests.

                  The thing that rubbed me about this, is that if SOP here isn't to look at those sorts of things to ensure a consistent test setup prior to running benchmarks, then all benchmark results are going to be suspect IMHO.
                  But is is a consistent setup. Getting a top measurement from before benchmarking is great if you're running a live desktop box, doing a comparison between disparate systems, etc. But that's not what this is. This is "Benchmark. Replace Kernel. Benchmark. Replace Kernel. Benchmark." If something is screwed up between benchmarks, it's due to the replacement kernel. Having a top number might give you more information about what part is causing the slowdown, but it doesn't remove the fact that there is a slowdown, which is all that the Phoromatic is designed to test. The fact it broke udev is secondary to the actual problem.

                  Again, Phoromatic is not a diagnosis tool, it is a benchmark. It is designed to tell you how fast, not why it is or isn't as fast as you're expecting.

                  Comment


                  • Yes, I agree that this is a bit large for such a minor thing.
                    Still within the merge window (or shortly after) then regressions like this are expected.

                    I wouldn't say stop warning us.. Back when I used an NVIDIA card, I used beta drivers and would appreciate the "heads-up" about a specific beta driver running slow.

                    I wouldn't expect so much hoopla over it though.
                    Maybe some simple graphs and enough testing to say "actually, yes, it is a problem with the drivers / software".

                    I say continue warning us - but just try to keep it to 1 page with a summary / enlarging summary pic showing some combined graphs.

                    Comment


                    • The actual reason for Michael's post

                      Michael seems to be RedHat's dog, Rover.

                      The RedHat developers, and in particular, Dave Airlie have come under intense scrutiny and criticism by core kernel developers, for their inept and ultimately damaging stupidity.

                      Michael, who is partly funded by RedHat, appears to have been asked to redress the balance, as it were. And the net result is a damaging hit-piece that comes straight off the front page of the News Of The World or National Enquirer.

                      But this is not the first time Michael has not been impartial, by any means. And it certainly won't be the last.

                      Hey Michael - Fetch that stick ! Good doggy !

                      Comment


                      • How about we keep the personal attacks away?

                        Comment


                        • Originally posted by gordboy View Post
                          Michael seems to be RedHat's dog, Rover.

                          The RedHat developers, and in particular, Dave Airlie have come under intense scrutiny and criticism by core kernel developers, for their inept and ultimately damaging stupidity.

                          Michael, who is partly funded by RedHat, appears to have been asked to redress the balance, as it were. And the net result is a damaging hit-piece that comes straight off the front page of the News Of The World or National Enquirer.

                          But this is not the first time Michael has not been impartial, by any means. And it certainly won't be the last.

                          Hey Michael - Fetch that stick ! Good doggy !
                          I am funded by Red Hat? Wow, since when? I didn't know.
                          Michael Larabel
                          http://www.michaellarabel.com/

                          Comment


                          • Originally posted by gordboy View Post
                            Michael seems to be RedHat's dog, Rover.

                            The RedHat developers, and in particular, Dave Airlie have come under intense scrutiny and criticism by core kernel developers, for their inept and ultimately damaging stupidity.

                            Michael, who is partly funded by RedHat, appears to have been asked to redress the balance, as it were. And the net result is a damaging hit-piece that comes straight off the front page of the News Of The World or National Enquirer.

                            But this is not the first time Michael has not been impartial, by any means. And it certainly won't be the last.

                            Hey Michael - Fetch that stick ! Good doggy !
                            and you have any evidence for your claims? Starting with Airlie being intensely critized to Redhats involvement? I am sure whoever does Phoronix accounting would be delighted.

                            Comment


                            • Originally posted by gordboy View Post
                              Michael seems to be RedHat's dog, Rover.

                              The RedHat developers, and in particular, Dave Airlie have come under intense scrutiny and criticism by core kernel developers, for their inept and ultimately damaging stupidity.

                              Michael, who is partly funded by RedHat, appears to have been asked to redress the balance, as it were. And the net result is a damaging hit-piece that comes straight off the front page of the News Of The World or National Enquirer.

                              But this is not the first time Michael has not been impartial, by any means. And it certainly won't be the last.

                              Hey Michael - Fetch that stick ! Good doggy !
                              get the conspiracy bibles out folks, the islamists are waging a war on any bar redhat

                              Comment


                              • Oh snap, our good friend gordboy is back


                                Whenever he posts the lulz follows.

                                Right now who gives a snap as the problem's been resolved. In the future I am sure Michael will give more detailed information about a problem that Phoromatic detects so that kernel devs will take a look at it.

                                I do think that michael should provide a meaningful bug report when an issue is detected and where the problem lies rather than blaring it out. The kernel devs are a hard working bunch that don't need to be screamed at like kids.

                                Comment

                                Working...
                                X