Greetings fellas,
I recently added server Load service and from what I got online values should be 0.9 for critical per core for 1m and 0.8 for warning etc etc for the rest.
Thing is, should these checks be like this? If I remember from my past job, is it wise to set values higher? Like CPU should be critical for number of CPU x2 . So like I can set it 1.8 for crit and 1.6 for warning for server with 1CPU for 1min check
Is that good or not? Can server CPU handle with ease bellow doubled load or are these merits on point?
Asking cause I'm trying to get better picture of state our servers from provided statistics and graphs.
Thanks in advance!
Load service - what are best values
-
- DevOps Engineer
- Posts: 19396
- Joined: Tue Nov 15, 2011 3:11 pm
- Location: Nagios Enterprises
- Contact:
Re: Load service - what are best values
This is really subjective depending on your needs, desires and application. A load > 1.0 per core means that a process will have to wait before being sent to the CPU for processing, but many applications can handle this just fine up to a certain point, it depends on the application and how long the wait ends up being.Shwele wrote:Is that good or not? Can server CPU handle with ease bellow doubled load or are these merits on point?
We have servers that can easily handle a short term load of 10 per core, they start to get a little sluggish but if it is not sustained, then it is not a problem.
Re: Load service - what are best values
I see well think I lack the skills to determine that... Thanks for explaining in great detail and confirming what I had on mind.scottwilkerson wrote:This is really subjective depending on your needs, desires and application. A load > 1.0 per core means that a process will have to wait before being sent to the CPU for processing, but many applications can handle this just fine up to a certain point, it depends on the application and how long the wait ends up being.Shwele wrote:Is that good or not? Can server CPU handle with ease bellow doubled load or are these merits on point?
We have servers that can easily handle a short term load of 10 per core, they start to get a little sluggish but if it is not sustained, then it is not a problem.
These are not sluggish from what I got on usage side. We have about 30 Drupal websites running with Aegir. We are providing non profits with this service in projects to use CiviCRM as database. Thing is, this server, has 1CPU and its on critical load for 6 days already, while it has about 3 websites active at some point in time. So just 95% of the time its passive and being alive, server has working load:
![Image](https://snag.gy/FlC3iS.jpg)
Uptime is 6 days, so server load is current.
And here is graph in the past week:
![Image](https://snag.gy/NMlWwP.jpg)
And for last 24 hours as well:
![Image](https://snag.gy/dPOHli.jpg)
Hope this spices up our discussion and tweaks!
-
- DevOps Engineer
- Posts: 19396
- Joined: Tue Nov 15, 2011 3:11 pm
- Location: Nagios Enterprises
- Contact:
Re: Load service - what are best values
I personally wouldn't be alarmed, if this was my server, based on what you describe, but again, it's up to your organizations specifications.
Re: Load service - what are best values
I seescottwilkerson wrote:I personally wouldn't be alarmed, if this was my server, based on what you describe, but again, it's up to your organizations specifications.
well I'm wondering then about basic setup, what is the best then load per CPU to setup?? Is there some global scale knowhow, like Load x2 per CPU or even higher clock checks?
I just want to have servers peachy and so I can know when they ought to have upgrades from personal experience.
Hopefully there will be someone else one of these days stumbling on this topic and providing their opinion as well.
Thanks for your effort and every future one (others too!)
-
- DevOps Engineer
- Posts: 19396
- Joined: Tue Nov 15, 2011 3:11 pm
- Location: Nagios Enterprises
- Contact:
Re: Load service - what are best values
I'll leave this open for other feedback but as I've mentioned before it's usually based on organization criteria, and load over 1.0 per CPU there starts to be a wait introduced before processes have direct access to resources, this could be CPU but also I/O wait can be a factor as well.