Loading ...
Sorry, an error occurred while loading the content.

Re: [solarisx86] Solaris 11.1 kstat cpu_info reports clock speed at half value

Expand Messages
  • John Martin
    ... The SMBIOS current speed field only reflects the processor speed at system boot. This is not a dynamic sampling of the processor speed for each call. With
    Message 1 of 13 , Dec 30, 2012
    • 0 Attachment
      On 12/29/12 21:27, Peter Schow wrote:
      > On Fri, Dec 28, 2012 at 09:17:21PM -0000, pcsol1996 wrote:
      > > Do I have a hardware/config issue with the X3-2 or is this just some
      > sort
      > > of weirdness with Solaris and psrinfo? I also checked my Ultra 40
      > running
      > > Solaris 11.1 and it reports the 3Ghz AMD CPUs at 3000Mhz.
      >
      > If you want to verify your CPU speeds, you can run smbios, a la:
      >
      > smbios -t 4
      >
      > which should tell you current and max.

      The SMBIOS current speed field only reflects the processor speed
      at system boot. This is not a dynamic sampling of the processor
      speed for each call.

      With P-states (Intel BIOSes may call this something like SpeedStep)
      the processor clock will jump between states based on system load
      and the top speed will change also (on my system this is labeled
      as "turbo").

      For example, on my i7 980X smbios reports 3333MHz for maximum
      and current speed. However, while typing this message powertop
      reports the clock has jumped between the bottom and top P-states
      and the top (turbo) value has changed from 3330 through 3457,
      with lots of steps in between.
    • Laurent Blume
      ... On any Solaris box, up to S10, that should give you the current clock speed and the supported speeds for each core: kstat
      Message 2 of 13 , Dec 31, 2012
      • 0 Attachment
        On 12/30/12 15:47, John Martin wrote:
        > The SMBIOS current speed field only reflects the processor speed
        > at system boot. This is not a dynamic sampling of the processor
        > speed for each call.
        >
        > With P-states (Intel BIOSes may call this something like SpeedStep)
        > the processor clock will jump between states based on system load
        > and the top speed will change also (on my system this is labeled
        > as "turbo").
        >
        > For example, on my i7 980X smbios reports 3333MHz for maximum
        > and current speed. However, while typing this message powertop
        > reports the clock has jumped between the bottom and top P-states
        > and the top (turbo) value has changed from 3330 through 3457,
        > with lots of steps in between.

        On any Solaris box, up to S10, that should give you the current clock
        speed and the supported speeds for each core:

        kstat 'cpu_info:::/current_clock_Hz|supported_frequencies_Hz/'

        It is dynamic and kstat (luckily for the quality of the metrics) does
        not use enough CPU power to make it jump out of the lowest state.

        I see no point in wanting to peg the CPU at the higher level. It would
        use more energy, require more cooling, and those days, probably would
        not yield any visible performance increase.

        Laurent
      • Bob Friesenhahn
        ... On my OpenIndiana Xeon E5 system which psrinfo -v reports that the clocks are 2.7GHZ, the above command reports that the cores are actually running at
        Message 3 of 13 , Dec 31, 2012
        • 0 Attachment
          On Mon, 31 Dec 2012, Laurent Blume wrote:
          >
          > On any Solaris box, up to S10, that should give you the current clock
          > speed and the supported speeds for each core:
          >
          > kstat 'cpu_info:::/current_clock_Hz|supported_frequencies_Hz/'
          >
          > It is dynamic and kstat (luckily for the quality of the metrics) does
          > not use enough CPU power to make it jump out of the lowest state.

          On my OpenIndiana Xeon E5 system which 'psrinfo -v' reports that the
          clocks are 2.7GHZ, the above command reports that the cores are
          actually running at 1,200,000,000 HZ, which is also the lowest
          supported frequency.

          The lower clocks should result in less coal being burned today.

          Bob
          --
          Bob Friesenhahn
          bfriesen@..., http://www.simplesystems.org/users/bfriesen/
          GraphicsMagick Maintainer, http://www.GraphicsMagick.org/
        • John Martin
          ... I hope I didn t imply this was the case. While powertop did show transitions between the highest (1596MHz) and lowest (3333MHz+) P-states, over 95% of the
          Message 4 of 13 , Dec 31, 2012
          • 0 Attachment
            On 12/31/12 04:32, Laurent Blume wrote:
            > On 12/30/12 15:47, John Martin wrote:
            > ...
            > > For example, on my i7 980X smbios reports 3333MHz for maximum
            > > and current speed. However, while typing this message powertop
            > > reports the clock has jumped between the bottom and top P-states
            > > and the top (turbo) value has changed from 3330 through 3457,
            > > with lots of steps in between.
            > ...
            > I see no point in wanting to peg the CPU at the higher level. It would
            > use more energy, require more cooling, and those days, probably would
            > not yield any visible performance increase.

            I hope I didn't imply this was the case. While powertop did show
            transitions between the highest (1596MHz) and lowest (3333MHz+)
            P-states, over 95% of the time was spent in the highest state.

            I would expect most users would want this behavior and not
            artificially force the processor to run with the fastest
            clock when there isn't sufficient load. If there are loads
            which don't transition the processor to the lowest P-state,
            we can look at the test case to see if it is a bug.
          • Laurent Blume
            ... I think it works rather well, but the documentation (for Solaris 10 and OpenSolaris around b111 time) was a little difficult to grasp when I set it up, and
            Message 5 of 13 , Jan 2, 2013
            • 0 Attachment
              On 01/01/13 01:02, John Martin wrote:
              > I hope I didn't imply this was the case. While powertop did show
              > transitions between the highest (1596MHz) and lowest (3333MHz+)
              > P-states, over 95% of the time was spent in the highest state.
              >
              > I would expect most users would want this behavior and not
              > artificially force the processor to run with the fastest
              > clock when there isn't sufficient load. If there are loads
              > which don't transition the processor to the lowest P-state,
              > we can look at the test case to see if it is a bug.

              I think it works rather well, but the documentation (for Solaris 10 and
              OpenSolaris around b111 time) was a little difficult to grasp when I set
              it up, and the default power.conf rather unhelpful. I've had a look at
              times since then, but nothing very thorough. There were also bugs at
              time on the CPU load report, but I think they've all been fixed by now?
              I've not looked at it in a while. I probably should check on S11 how
              it's doing.

              Laurent
            • pcsol1996
              ... Thanks for all the replies, folks.
              Message 6 of 13 , Jan 2, 2013
              • 0 Attachment
                --- In solarisx86@yahoogroups.com, Peter Schow <pschow@...> wrote:
                >
                > On Fri, Dec 28, 2012 at 09:17:21PM -0000, pcsol1996 wrote:
                > > Do I have a hardware/config issue with the X3-2 or is this just some sort
                > > of weirdness with Solaris and psrinfo? I also checked my Ultra 40 running
                > > Solaris 11.1 and it reports the 3Ghz AMD CPUs at 3000Mhz.
                >
                > If you want to verify your CPU speeds, you can run smbios, a la:
                >
                > smbios -t 4
                >
                > which should tell you current and max.
                >
                Thanks for all the replies, folks.
              • Marc Lobelle
                Dear all, First my best wishes for 2013! I just noticed a surprising disk space leak on my solaris 11 notebook (A HP Elitebook 2540 with a SSD drive) The drive
                Message 7 of 13 , Jan 7, 2013
                • 0 Attachment
                  Dear all,

                  First my best wishes for 2013!

                  I just noticed a surprising disk space leak on my solaris 11 notebook (A HP Elitebook 2540 with a
                  SSD drive)
                  The drive is partitiones in a Widows 7 partition of about 60 Gig and a solaris partition of about
                  150 Gig. When I tried to update solaris 11, as I am regularly proposed to do, it could not do it
                  because there was not enough disk space. So I freed all I could miss and nowthere is 9 free gigs so
                  that I shalle be able to do the update However if I do df -k, I read;
                  -[2]-~# df -k
                  Système de fichiers blocs de 1024 Utilisé Disponible Capacité Monté sur
                  rpool/ROOT/solaris-4 143474688 15044076 9966849 61% /
                  /devices 0 0 0 0% /devices
                  /dev 0 0 0 0% /dev
                  ctfs 0 0 0 0% /system/contract
                  proc 0 0 0 0% /proc
                  mnttab 0 0 0 0% /etc/mnttab
                  swap 3864148 692 3863456 1% /system/volatile
                  objfs 0 0 0 0% /system/object
                  sharefs 0 0 0 0% /etc/dfs/sharetab
                  /usr/lib/libc/libc_hwcap1.so.1
                  25010925 15044076 9966849 61% /lib/libc.so.1
                  fd 0 0 0 0% /dev/fd
                  swap 3863508 52 3863456 1% /tmp
                  rpool/export 143474688 32 9966849 1% /export
                  rpool/export/home 143474688 32 9966849 1% /export/home
                  rpool/export/home/ml 143474688 14011760 9966849 59% /export/home/ml
                  rpool 143474688 98 9966849 1% /rpool
                  /export/home/ml 23978609 14011760 9966849 59% /home/ml

                  If I understand correctly the disk has 143 gigs, of which my personal directories use 14 Gigs and
                  the system 15 gigs

                  Where are the missing 110 gigs ??

                  Does anybody understand this ?

                  Thanks

                  Marc
                • Laurent Blume
                  ... Yes! Happy new year to all! ... [snip] ... Mostly, with zfs, df has become largely irrelevant in estimating disk space use. What you need to do is «beadm
                  Message 8 of 13 , Jan 7, 2013
                  • 0 Attachment
                    On 01/07/13 12:19, Marc Lobelle wrote:
                    > Dear all,
                    >
                    > First my best wishes for 2013!

                    Yes! Happy new year to all!

                    > I just noticed a surprising disk space leak on my solaris 11 notebook (A HP Elitebook 2540 with a
                    > SSD drive)
                    > The drive is partitiones in a Widows 7 partition of about 60 Gig and a solaris partition of about
                    > 150 Gig. When I tried to update solaris 11, as I am regularly proposed to do, it could not do it
                    > because there was not enough disk space. So I freed all I could miss and nowthere is 9 free gigs so
                    > that I shalle be able to do the update However if I do df -k, I read;
                    > -[2]-~# df -k

                    [snip]

                    > If I understand correctly the disk has 143 gigs, of which my personal directories use 14 Gigs and
                    > the system 15 gigs
                    >
                    > Where are the missing 110 gigs ??
                    >
                    > Does anybody understand this ?

                    Mostly, with zfs, df has become largely irrelevant in estimating disk
                    space use.

                    What you need to do is «beadm list» and »«zfs list -t all -r rpool».
                    The first will show you if there are old BE's that are not needed
                    anymore (the space they actually use is always bigger than what the
                    command says). The second will show you all the datasets on rpool,
                    mounted or not, and their snapshots.
                    There are probably some of those that you can destroy.

                    Hope this helps,

                    Laurent
                  • Laurent Blume
                    ... The usual rule of thumb is to keep at least one working previous environment, just in case. So you can remove the older ones, yes. Also note that when you
                    Message 9 of 13 , Jan 9, 2013
                    • 0 Attachment
                      On 08/01/13 00:06, Marc Lobelle wrote:
                      > there are 5:
                      > solaris: 91.78 M dated 2010-12-08
                      > solaris-1 : 54,69 M dated 2011-02-03
                      > solaris-2: 34,13 M dated 2011-0203
                      > solaris-3: 691,0K dated 2012-11-0
                      >
                      > may I safely destroy these old boot environments?

                      The usual rule of thumb is to keep at least one working previous
                      environment, just in case. So you can remove the older ones, yes. Also
                      note that when you upgrade the rpool's zpool or zfs versions, previous
                      BE's that don't support it will be unusable, so they can be destroyed.

                      > solaris-4: NR / 110,85 G 2012-11-06, which is obviously the one I use,
                      > but the partition is larger: where are the missing 33G ?

                      The calculation for space use are very confusing. Space use that is
                      shared between datasets is not shown. So, check the free space before
                      and after you destroy the oldest BE's: you will see it grows more than
                      expected.

                      > I see first
                      > rpool used: 127G, avail 9,49G, refer 98K mounted on /rpool
                      > rpool/root 110G 9,49G 31K legacy
                      >
                      > Next I see mounted rpools named after the BEs, plus unmounted ones called
                      > rpool/ROOT/solaris-4@install
                      > rpool/ROOT/solaris-4@2012-02-03-08-...
                      > rpool/ROOT/solaris-4@2012-02-03-09-...
                      > rpool/ROOT/solaris-4@2012-11-06-10-...
                      > rpool/ROOT/solaris-4@2012-11-06-11-...
                      > What is in there and what is their use The used space in these rpools is
                      > small but the last column , called refer, says 3,34G, 97,7G, 98,8G,
                      > 14,3G, 14,3G. What does that column mean
                      > and finally dump, export export/home, export/home/ml and swap
                      > What should I do to recover disk space ?

                      Those snapshots are probably those using most of the space. If you don't
                      need them anymore, you can remove them. The same thing about shared
                      space applies. «Refer» is how much usage they refer to (some of it can
                      be shared with others), «used» is the space that is specific to them. So
                      destroying them will recover some value between the two.

                      > du -ks / says 32.149.02 = 32,15G + 9,49 free = 41,64 so where are the
                      > missing 90 GB?

                      In the snapshots, du cannot account for them.

                      Laurent
                    Your message has been successfully submitted and would be delivered to recipients shortly.