Sytem Load

edited September 2016 in Features and Ideas
Why is it that the system load fluctuates so drastically for no apparent reason?
System is idle overall, 99.5% CPU idle, no major IP activity going on, no I/O, and the 1 minute load fluctuates from 0.02 to 0.5!! During the day it's like 0.40 for 15 min average.

Busy neighbors on the same hypervisor as my FreeBSD VPS is?

Comments

  • There is so many possible reasons for this... what do you do on your server whats happening on it could be updates,crons,backups,scripts and like 99 other things...

    And all of them relate to your server and not your neighbours...
  • edited October 2016
    System load can be tricky to work out at times. It's simply a measure of "stuff waiting to run" - and the delay could be caused by many reasons.

    My main home server (bare metal FreeBSD) is currently showing this, which may appear illogical:

    last pid: 28167; load averages: 0.22, 0.24, 0.19 up 0+13:58:29 17:54:27 132 processes: 2 running, 114 sleeping, 16 waiting CPU 0: 0.0% user, 0.0% nice, 0.0% system, 0.0% interrupt, 100% idle CPU 1: 0.0% user, 0.0% nice, 0.0% system, 0.0% interrupt, 100% idle Mem: 17M Active, 1615M Inact, 661M Wired, 32K Cache, 410M Buf, 1554M Free Swap: 4096M Total, 4096M Free

  • virami:
    There are, but as I was saying, in my case nothing is happening when I monitor it, the system is idle.
    That's totally not true that the load on a guest OS is not related to the host machine/hypervisor. I've had once a Proxmox hypervisor which started having HDD read/write issue, as a result the VMs hosted on it were slow as well, load was about 400 at times :)

    jamie:
    I understand it's tricky, however it's not normal to fluctuate that much, and the other thing is that 'before' it wasn't like that.
    Also, in your example you aren't watching for the disk I/O, which also is reflected on the load. Do a "systat -vmstat 1" as you'll see the full picture.
  • edited November 2016
    This actually isn't an issue with VULTR, it's an issue with FreeBSD's load reporting. There's a few topics on the FreeBSD mailing lists and forums about the high load reports on brand new/idle FreeBSD servers. There's a bug report that has been "In Progress" for quite some time on it.

    For example if you spin up a new VULTR (or any other vps provide for that matter, I've seen it happen on a number of other providers) FreeBSD VPS the load can go as high as 0.50 with nothing going on (idle cpu and pretty much idle disk and low interrupt rate). When my loads were reporting a 2.0 load average, VULTR's panel cpu usage was showing under %5 and very little disk usage.

    The fix I use (which I got from some of the mailing list and forums topics about the issue) is I add these two lines to /boot/loader.conf:

    kern.eventtimer.periodic = 1
    kern.hz = 100

    then reboot and my FreeBSD VPS loads are much much lower.

    Here's my load now (which is a big difference from reporting as high as 1.50-2):
    root@vultr-fbsd:~ # uptime
    9:13PM up 19:21, 1 users, load averages: 0.01, 0.02, 0.00

    Maybe some other FreeBSD users can chime in with their experiences :-). Besides the load thing, I absolutely love using my FreeBSD servers.
  • That was it! My load is now much lower.

    Actually, I had already kern.hz=100 with the comment "reduce idle CPU utilization in virtual environment, not good for Asterisk IVR and such". I didn't have the periodic parameter though.

    Thanks a bunch!
  • No problem, so glad that worked for you! :-). You used to only need the kern.hz part but since FreeBSD moved to tickless mode it requires you to set the eventtimer before the kern.hz setting takes effect.
Sign In or Register to comment.