11

Recently I discovered that our server does not utilize all 80 threads anymore in the system. It looks like if 16 cores are always idle, despite the high system load.

Its a Dell powerEdge R900 server, with 4 sockets, 4 times a 10-core Xeon. So 40 cores, with HT its 80 threads. (Intel(R) Xeon(R) CPU E7- 4850 @ 2.00GHz). System memory is 512GB Running Ubuntu 14.04.1 LTS. I haven't rebooted the server yet, I was hoping to avoid this.

uname -a
Linux assembly 3.13.0-35-generic #62-Ubuntu SMP Fri Aug 15 01:58:42 UTC 2014 x86_64 x86_64 x86_64 GNU/Linux

I'v check the following:

Temperature measured with i7z: (it cannot display 4 sockets

Cpu speed from cpuinfo 1994.00Mhz
True Frequency (without accounting Turbo) 1994 MHz

Socket [0] - [physical cores=10, logical cores=20, max online cores ever=10]
  CPU Multiplier 15x || Bus clock frequency (BCLK) 132.93 MHz
  TURBO ENABLED on 10 Cores, Hyper Threading ON
  Max Frequency without considering Turbo 2126.93 MHz (132.93 x [16])
  Max TURBO Multiplier (if Enabled) with 1/2/3/4/5/6 cores is  0x/0x/0x/0x/0x/0x
  Real Current Frequency 1994.02 MHz (Max of below)
        Core [core-id]  :Actual Freq (Mult.)      C0%   Halt(C1)%  C3 %   C6 %  Temp
        Core 1 [1]:       1994.01 (15.00x)       100       0       0       0    75
        Core 2 [5]:       1994.00 (15.00x)       100       0       0       0    77
        Core 3 [9]:       1994.02 (15.00x)       100       0       0       0    76
        Core 4 [13]:      1994.00 (15.00x)       100       0       0       0    77
        Core 5 [17]:      1994.00 (15.00x)       100       0       0       0    77
        Core 6 [21]:      1994.00 (15.00x)      97.7    0.404      0    1.86    77
        Core 7 [25]:      1994.00 (15.00x)      94.5       0       1    5.27    77
        Core 8 [29]:      1994.00 (15.00x)       100       0       0       0    76
        Core 9 [33]:      1994.00 (15.00x)      99.8       0       1       1    75
        Core 10 [37]:     1994.00 (15.00x)       100       0       0       0    73
  Max Frequency without considering Turbo 2126.93 MHz (132.93 x [16])
  Max TURBO Multiplier (if Enabled) with 1/2/3/4/5/6 cores is  0x/0x/0x/0x/0x/0x
  Real Current Frequency 1994.02 MHz (Max of below)
        Core [core-id]  :Actual Freq (Mult.)      C0%   Halt(C1)%  C3 %   C6 %  Temp
        Core 1 [1]:       1994.02 (15.00x)       100       0       0       0    74
        Core 2 [5]:       1994.00 (15.00x)       100       0       0       0    76
        Core 3 [9]:       1994.02 (15.00x)       100       0       0       0    76
        Core 4 [13]:      1994.00 (15.00x)       100       0       0       0    77
        Core 5 [17]:      1994.00 (15.00x)       100       0       0       0    76
        Core 6 [21]:      1994.00 (15.00x)        97       0       1    2.43    77
        Core 7 [25]:      1994.00 (15.00x)      92.9       0       1    6.81    77
C0 = Processor running without halting00x)       100       0       0       0    75
C1 = Processor running with halts (States >C0 are power saver)     1       1    75
C3 = Cores running with PLL turned off and core cache turned off   0       0    73
C6 = Everything in C3 + core state saved to last level cache
  Above values in table are in percentage over the last 1 sec
[core-id] refers to core-id number in /proc/cpuinfo
'Garbage Values' message printed when garbage values are read
  Ctrl+C to exit

Idle: Last 16 cores are all 100 % idle:

mpstat -p ALL 1:
Average:     CPU    %usr   %nice    %sys %iowait    %irq   %soft  %steal  %guest  %gnice   %idle
Average:     all   70.69    0.00    0.70    0.00    0.00    0.00    0.00    0.00    0.00   28.61
Average:       0   92.93    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    7.07
Average:       1   94.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    6.00
Average:       2  100.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00
Average:       3   83.33    0.00    2.08    0.00    0.00    0.00    0.00    0.00    0.00   14.58
Average:       4  100.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00
Average:       5  100.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00
..........................................................
Average:      64    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      65    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      66    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      67    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      68    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      69    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      70    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      71    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      72    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      73    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      74    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      75    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      76    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      77    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      78    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      79    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00

Top load numbers:

top - 17:41:48 up 35 days,  6:28, 15 users,  load average: 77.69, 70.48, 62.73
Tasks: 1327 total,  44 running, 1281 sleeping,   2 stopped,   0 zombie
%Cpu(s): 63.7 us, 13.6 sy,  0.0 ni, 22.3 id,  0.2 wa,  0.0 hi,  0.2 si,  0.0 st
KiB Mem:  52837942+total, 52553190+used,  2847524 free,   535660 buffers
KiB Swap: 78124032 total,  2105608 used, 76018416 free. 40637328+cached Mem

Sometimes the idle % is not 100 anymore, but a bit less, as you can see here, but mostly it stays at 100% idle.

Average:     CPU    %usr   %nice    %sys %iowait    %irq   %soft  %steal  %guest  %gnice   %idle
Average:     all   70.69    0.00    0.70    0.00    0.00    0.00    0.00    0.00    0.00   28.61
Average:      64    0.13    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00   99.87
Average:      65    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      66    0.00    0.00    2.63    0.00    0.00    0.00    0.00    0.00    0.00   97.37
Average:      67    0.00    0.00    0.13    0.13    0.00    0.00    0.00    0.00    0.00   99.75
Average:      68    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      69    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      70    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      71    0.00    0.00    0.12    0.00    0.00    0.00    0.00    0.00    0.00   99.88
Average:      72    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      73    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      74    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      75    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      76    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      77    0.00    0.00    0.13    0.00    0.00    0.00    0.00    0.00    0.00   99.87
Average:      78    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00
Average:      79    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00    0.00  100.00

I also ran this command to make sure they are all online:

for COUNT in `seq 01 79`;do echo 1 > /sys/devices/system/cpu/cpu${COUNT}/online; 

With the HTop program I can visualize a bar of cpu usage per thread, and is see 64 bars filled and 16 empty ones (the last 16).

When I try to start a process at a core >63 is also fails doing so:

root@server:~# taskset -c 63 time
Usage: time [-apvV] [-f format] [-o file] [--append] [--verbose]
       [--portability] [--format=format] [--output=file] [--version]
       [--quiet] [--help] command [arg...]
root@server:~# taskset -c 64 time
taskset: failed to set pid 0's affinity: Invalid argument
root@server:~# taskset -c 65 time
taskset: failed to set pid 0's affinity: Invalid argument

Related thread: https://askubuntu.com/questions/536541/ubuntu-uses-only-2-out-of-4-processor-cores

EDIT: It turns out that that the cores are shutdown on the fly, but do not startup properly. It seems that there are processes running on these unavailable cores, but its impossible to start any new process on them. According to the dmesg log, cores are disabled and enabled quickly after one another. I have to say that it was the intention to shutdown these cores, so we disabled this 'feature'. DMESG example log:

[Mon Jan 12 12:42:40 2015] kvm: disabling virtualization on CPU79
[Mon Jan 12 12:42:40 2015] smpboot: CPU 79 is now offline
....
[Mon Jan 12 12:43:12 2015] smpboot: Booting Node 0 Processor 79 APIC 0xf3
[Mon Jan 12 12:43:12 2015] kvm: enabling virtualization on CPU79

We enabling/disabling cores via:

for COUNT in `seq 64 79`;do echo 1 > /sys/devices/system/cpu/cpu${COUNT}/online;done

We never linked this commands with our 16 unavailable cores, since normally the commands above work properly. (we also tried disabling powermanager, but this did not help)

5
  • 1
    Why you or your company didnt go with Ubuntu server rather than Ubuntu for normal PCs?
    – XperianX
    Jan 15, 2015 at 9:00
  • 1
    It's not impossible that 16 of your cores are actually being used to MAX capacity, leaving 16 virtual cores with no usable margin.
    – anonymous2
    Jun 22, 2016 at 12:25
  • 1
    Are all 8 cores (16 virtual) on one physical CPU, or are they spread out across all 4 sockets? If they're located on one physical CPU, is it possible that those cores have failed? Jun 29, 2016 at 15:57
  • 2
    I am wondering if somewhere in the system there is code which represents the CPU affinity mask with just 64 bits preventing it from allowing processes to utilise the last 16 cores.
    – kasperd
    Nov 26, 2016 at 17:54
  • Given that 80-16 = 64, it seems to me like there's some artificial limit somewhere on the maximum number of cores to use. Possibly whatever software you're running is set to a max of 64 cores, or possibly it can't generate more than 64 threads.
    – Clonkex
    Nov 23, 2017 at 22:52

1 Answer 1

1

Not all programs can use multiple threads. php is one for example. If one php process needs a lot of CPU, only 1 CPU will max out. And the others will be idle.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .