Author Topic: Decipher System Monitor load average  (Read 8986 times)

Offline c4rv

  • Decent Indexer
  • ***
  • Posts: 75
  • Helpful: +3/-0
Decipher System Monitor load average
« on: 2014-08-05, 04:05:57 AM »
New to managing linux server, trying to decipher was load average mean in System Monitor

Showing for last 1, 5, 15 mintues - 17.42, 15.25, 13.31


Is this good / bad. Is there anything I need to tune ?

Thanks,

Offline Wally73

  • Overlord
  • ******
  • Posts: 260
  • Helpful: +28/-1
  • i'm nuts
Re: Decipher System Monitor load average
« Reply #1 on: 2014-08-05, 04:43:45 AM »
that is very high load

those are mine load average: 0.84, 0.62, 0.69 everything below 1 is ok after 2.x system system responds very slowly

what are you running?

Offline c4rv

  • Decent Indexer
  • ***
  • Posts: 75
  • Helpful: +3/-0
Re: Decipher System Monitor load average
« Reply #2 on: 2014-08-05, 04:52:31 AM »
I've using hyper-V which is probably confusing things.

Host is W2K12 with xeon v3-1230 (quad core with hyper threading) and 32GB of RAM. nZEDb instance is 2 virtual CPU, 25GB RAM and dedicated (pass through) SSD.

Offline c4rv

  • Decent Indexer
  • ***
  • Posts: 75
  • Helpful: +3/-0
Re: Decipher System Monitor load average
« Reply #3 on: 2014-08-05, 05:22:46 AM »
I've shutdown my minecraft server and doubled vCPU. Give it a couple of hours and see what happens.

Offline kevin123

  • Overlord
  • ******
  • Posts: 456
  • Helpful: +49/-0
Re: Decipher System Monitor load average
« Reply #4 on: 2014-08-05, 06:07:41 AM »
Load does not correspond exactly to CPU usage, it's for the overall system. If you have 1 activity holding up something else your load can skyrocket, yet your CPU can be idle.

An example is if you have a program that has tons of calculations to do on the CPU, but the data is coming from a SD card, the SD card can only put out 1% of the bandwidth the CPU can process, then your CPU is @ 1% but your load keeps rising, since the program has already queued many instructions but is waiting on the SD card to be able to finish them.

Offline c4rv

  • Decent Indexer
  • ***
  • Posts: 75
  • Helpful: +3/-0
Re: Decipher System Monitor load average
« Reply #5 on: 2014-08-05, 06:47:42 AM »
I see what you are saying, but over system performance should still be OK as CPU can service other threads while waiting for SD card.

Been an hour, I am guessing the CPU was overloaded as search were taking up to 30 secs and the screen was very sluggish at times. Since the change, its a littler snappier and now search are down to 10 secs and load is showing as under 10.

Offline NZBmets

  • Prolific Indexer
  • ****
  • Posts: 260
  • Helpful: +11/-0
Re: Decipher System Monitor load average
« Reply #6 on: 2014-08-05, 09:03:55 AM »
There is also often a large variance between actual load and what you see as a load measurement on a VM, most especially when it is Hyper-V as hypervisor and *nix on the VM.

Offline kevin123

  • Overlord
  • ******
  • Posts: 456
  • Helpful: +49/-0
Re: Decipher System Monitor load average
« Reply #7 on: 2014-08-05, 09:50:28 AM »
I see what you are saying, but over system performance should still be OK as CPU can service other threads while waiting for SD card.

Yep, that's why I wrote load doesn't correspond exactly to CPU usage :)
Another example I can give, archer} had an NFS that was having issues and something on his server was trying to write to the NFS but couldn't, so the load kept climbing, it was in the 100's yet the CPU was almost idle. The server was running fine, and as you wrote it was able to perform other tasks easily.

Offline tcpsyn

  • Newbie
  • *
  • Posts: 3
  • Helpful: +0/-0
Re: Decipher System Monitor load average
« Reply #8 on: 2014-09-06, 02:59:28 PM »
Generally, what you are are looking for is no higher than the number of CPUs.

A load of 1 is maxed on a single core cpu.
On a quad core, 4 is maxed.

The load is split into three... avg load after 1 minute, 5 minutes, 15 minutes.

Like has been mentioned, it is more complex and io/network play a part... but generally speaking, if your load is greater than your number of cores... there is a problem that needs to be identified. In my experience, it's usually boils down to disk performance.

Offline kingcat

  • Junior Indexer
  • **
  • Posts: 20
  • Helpful: +5/-0
  • NZB.Cat
    • NZB.Cat
Re: Decipher System Monitor load average
« Reply #9 on: 2015-02-10, 08:36:47 AM »
In linux, load average means nothing more or less than how many processes were waiting to run in a given time period (1 minute, 5 minutes, 15 minutes).

As others have said, a good rule of thumb is a load <= the amount of cores available. i.e. 6 cores, load of 6,6,6 would be fine (and also awesome). Given that scenario it means 6 processes are always waiting to run, on a 1 core box this would be a huge problem but with 6 cores, each of those processes are likely next up in the queue to run.

Since it has nothing to do with specific subsystems (ram, cpu, io, etc) it could be any or all of them that contribute to your load average. Using utilities like iostat and top (or preferably htop) will help you narrow down whats causing your unduly high load.


99% Completion, 2000 Days Retention
https://nzb.cat