This discussion is archived
2 Replies Latest reply: May 5, 2010 6:09 AM by 807559 RSS

CPU Usage

807559 Newbie
Currently Being Moderated
I have a question posed to me by an Oracle DBA who raised concerns that 2 virtual CPU cores are steady at about 94% idle while the 4 main cores are about 85% idle. The "top" command says the system has a load of about 3 to 4 and is about 50% idle. I figured the system would onlu\y use what it needed but the DBA insisted the idle times should be load balanced to have exactly the same idle times across all 6 CPUs. I have added some output to view. My question is this. Is the 95% idle time on the 2 virtual cores considered normal? If not, what could I do to better balance the CPU load?

TOP OUTPUT
last pid: 15493; load averages: 2.72, 3.14, 3.42 16:03:40
1371 processes:1363 sleeping, 2 running, 6 on cpu
CPU states: 52.7% idle, 38.0% user, 8.7% kernel, 0.6% iowait, 0.0% swap
Memory: 40G real, 16G free, 21G swap in use, 43G swap free

PID USERNAME LWP PRI NICE SIZE RES STATE TIME CPU COMMAND
12879 oracle 1 0 11 64M 44M run 31:37 16.31% frmweb
398 root 1 29 0 1968K 904K cpu/0 20.4H 2.53% praudit
396 root 2 19 0 2688K 1432K cpu/1 812:40 1.64% snarecore
215 root 30 59 0 4024K 3024K sleep 337:27 0.63% nscd
19255 oracle 1 26 11 56M 37M sleep 1:08 0.44% frmweb
11688 oracle 1 19 11 42M 24M sleep 0:05 0.42% frmweb
15179 oracle 1 26 11 40M 22M sleep 0:08 0.42% frmweb
698 oracle 436 26 11 523M 370M sleep 8:42 0.41% java
12627 oracle 1 39 0 4144K 2480K sleep 1:00 0.39% top
15275 root 1 39 0 4096K 2512K cpu/16 0:04 0.39% top
11509 oracle 1 0 11 52M 34M run 1:13 0.37% frmweb
9445 oracle 1 16 11 61M 42M sleep 1:15 0.31% frmweb
694 oracle 433 26 11 459M 306M sleep 9:51 0.29% java
27107 oracle 1 26 11 47M 29M sleep 0:50 0.29% frmweb
12787 oracle 1 26 11 52M 33M sleep 0:29 0.28% frmweb

TROVIASPROD4:/ #psrinfo
0 on-line since 04/11/2010 15:38:19
1 on-line since 04/11/2010 15:38:19
2 on-line since 04/11/2010 15:38:19
3 on-line since 04/11/2010 15:38:14
16 on-line since 04/22/2010 08:55:13
17 on-line since 04/22/2010 08:55:20
18 off-line since 04/11/2010 15:38:46
19 off-line since 04/11/2010 15:38:46

TROVIASPROD4:/ #mpstat -p
CPU minf mjf xcal intr ithr csw icsw migr smtx srw syscl usr sys wt idl set
0 79 0 1567 1869 253 1679 130 94 606 1 573 11 4 1 84 0
1 84 0 1557 1833 1 1621 138 99 543 1 267 11 3 1 85 0
2 83 0 1712 337 237 1676 111 94 683 1 742 12 4 1 84 0
3 92 0 1422 435 216 1628 134 94 602 1 454 11 3 0 85 0
16 33 0 809 50 18 720 30 43 247 0 2111 4 1 0 *94* 0 <--------idle time in question
17 37 0 842 32 1 691 31 45 233 0 1982 5 1 0 *94* 0 <--------idle time in question
  • 1. Re: CPU Usage
    user4994457 Newbie
    Currently Being Moderated
    TAdriver wrote:
    I have a question posed to me by an Oracle DBA who raised concerns that 2 virtual CPU cores are steady at about 94% idle while the 4 main cores are about 85% idle.
    What do you mean by a "virtual core" and a "main core"? The items you are looking at are "virtual CPUs

    What is the hardware in question? Can you show "psrinfo -vp" output?
    The "top" command says the system has a load of about 3 to 4 and is about 50% idle. I figured the system would onlu\y use what it needed but the DBA insisted the idle times should be load balanced to have exactly the same idle times across all 6 CPUs.
    Depends.
    Interrupts are not sent to all CPUs, so that can cause some differences. Processor affinity could affect things short term, but it shouldn't be consistent. Also, you've disabled two virtual CPUs. Depending on the relationship, some CPU cores might not be loaded the same way.

    I have added some output to view. My question is this. Is the 95% idle time on the 2 virtual cores considered normal? If not, what could I do to better balance the CPU load?
    >
    TOP OUTPUT
    last pid: 15493; load averages: 2.72, 3.14, 3.42 16:03:40
    1371 processes:1363 sleeping, 2 running, 6 on cpu
    CPU states: 52.7% idle, 38.0% user, 8.7% kernel, 0.6% iowait, 0.0% swap
    Memory: 40G real, 16G free, 21G swap in use, 43G swap free

    PID USERNAME LWP PRI NICE SIZE RES STATE TIME CPU COMMAND
    12879 oracle 1 0 11 64M 44M run 31:37 16.31% frmweb
    398 root 1 29 0 1968K 904K cpu/0 20.4H 2.53% praudit
    396 root 2 19 0 2688K 1432K cpu/1 812:40 1.64% snarecore
    215 root 30 59 0 4024K 3024K sleep 337:27 0.63% nscd
    19255 oracle 1 26 11 56M 37M sleep 1:08 0.44% frmweb
    11688 oracle 1 19 11 42M 24M sleep 0:05 0.42% frmweb
    15179 oracle 1 26 11 40M 22M sleep 0:08 0.42% frmweb
    698 oracle 436 26 11 523M 370M sleep 8:42 0.41% java
    12627 oracle 1 39 0 4144K 2480K sleep 1:00 0.39% top
    15275 root 1 39 0 4096K 2512K cpu/16 0:04 0.39% top
    11509 oracle 1 0 11 52M 34M run 1:13 0.37% frmweb
    9445 oracle 1 16 11 61M 42M sleep 1:15 0.31% frmweb
    694 oracle 433 26 11 459M 306M sleep 9:51 0.29% java
    27107 oracle 1 26 11 47M 29M sleep 0:50 0.29% frmweb
    12787 oracle 1 26 11 52M 33M sleep 0:29 0.28% frmweb

    TROVIASPROD4:/ #psrinfo
    0 on-line since 04/11/2010 15:38:19
    1 on-line since 04/11/2010 15:38:19
    2 on-line since 04/11/2010 15:38:19
    3 on-line since 04/11/2010 15:38:14
    16 on-line since 04/22/2010 08:55:13
    17 on-line since 04/22/2010 08:55:20
    18 off-line since 04/11/2010 15:38:46
    19 off-line since 04/11/2010 15:38:46

    TROVIASPROD4:/ #mpstat -p
    CPU minf mjf xcal intr ithr csw icsw migr smtx srw syscl usr sys wt idl set
    0 79 0 1567 1869 253 1679 130 94 606 1 573 11 4 1 84 0
    1 84 0 1557 1833 1 1621 138 99 543 1 267 11 3 1 85 0
    2 83 0 1712 337 237 1676 111 94 683 1 742 12 4 1 84 0
    3 92 0 1422 435 216 1628 134 94 602 1 454 11 3 0 85 0
    16 33 0 809 50 18 720 30 43 247 0 2111 4 1 0 *94* 0 <--------idle time in question
    17 37 0 842 32 1 691 31 45 233 0 1982 5 1 0 *94* 0 <--------idle time in question
    Instead of disabling 18 and 19, I might try disabling 3 and 19 and see what the response looks like.

    --
    Darren
  • 2. Re: CPU Usage
    807559 Newbie
    Currently Being Moderated
    Hey Darren, thanks for clearing up my misunderstanding of virtual cores. I see what you are talking about. I have set my startup script to ensure processors 0, 1, 2, 16, 17 and 18 are brought on-line and 3 and 19 are set to off-line. The business need with licensing is the reason we have 2 CPUs in the off-line state. Of course, I'll have to wait till the next available reboot time for this but at least it should be as solid as it can get from the admin side.