whm load averages

[jbngar]

Hey guys

I understand the 3 'load averages' figures in the top right corner of the whm control panel reflect the CPU load for the last minute, five minutes and fifteen minutes, with 1 being 100% utilisation.

I also understand it is complicated by the number of cores available - if you have 2 cores you would half the figure. So a regular load average of 0.1 on a 2 core VPS should mean it is running at 0.05 = 5% of its capability - and probably can cope with a significant amount more.

Have I got this right, or is it further complicated by the amount of CPU resources allocated to the VPS plans?

My reason for asking is for both interest sake and also to ascertain a prudent number of clients to put on my VPS without straining resources.

Obviously running the CPUs at 100% all the time is not good - is there an industry standard load average range that is considered wise?

Cheers
jbng :-)

[Aaron]

Hi jbngar,

This article on our US division site might be able to help with this: Server Loads Explained

Let me know if covers what you are looking for.

Originally Posted by jbngar

Hey guys

I understand the 3 'load averages' figures in the top right corner of the whm control panel reflect the CPU load for the last minute, five minutes and fifteen minutes, with 1 being 100% utilisation.

I also understand it is complicated by the number of cores available - if you have 2 cores you would half the figure. So a regular load average of 0.1 on a 2 core VPS should mean it is running at 0.05 = 5% of its capability - and probably can cope with a significant amount more.

Have I got this right, or is it further complicated by the amount of CPU resources allocated to the VPS plans?

My reason for asking is for both interest sake and also to ascertain a prudent number of clients to put on my VPS without straining resources.

Obviously running the CPUs at 100% all the time is not good - is there an industry standard load average range that is considered wise?

Cheers
jbng :-)

[jbngar]

Thanks Aaron

That's helpful information, but it seems to apply more if you had a dedicated server that wasn't split into VPSs. My questions actually lie more with how the CPU usage is divided on the VPS hardware.

You list on your site that your servers are dual quad core processors (= 8 cores), with 32GB of RAM.

If everyone had the same 512 VPS plan I have, that would split the server into 64 accounts (32GB/512MB = 64). So those 64 accounts would be sharing the 8 cores, leaving each account with a total of 1/8th of a core each. But you state that the 512 plans get access to 2 cores, so I guess that means 1/16th of two cores.

So if the two CPUs run at 2Ghz, does that mean my VPS slice is the equivalent of dual 125Mhz processors (2Ghz/8) and I can't utilise more than that? Or does it work more like virtuozzo in that if no-one else is using those two cores I get full access to them, but if there's heavy load involved I don't go *below* that?

Either way, is my load average figure based on my 1/16th of a core, or does it reflect the total core - ie if everyone's accounts on the node were maxed out, would a load average of 0.0625 (1/16th of 1) mean my server was using all its CPU resources?

Again, the main objectives for me are firstly to make sure that I'm not ignorantly overestimating my capacity and having my clients suffer for it, and secondly to understand more about Xen and its shared CPU usage.

If my VPS is performing really well now (which it appears to be, thanks!) I'd like to know whether it's only doing that because hardly anyone else is on the node and I'm actually getting close to 2 full cores of CPU usage rather than just my 1/8th allocation. In which case there may be performance degradation as more clients are added to the node and they start utilising their CPU allocation.

On the other hand, if my VPS is performing just great on my 1/8th allocation while the node is maxed out with other clients who cannot steal my resources, then I know I can confidently operate within my dedicated resource budget.

Does that make sense? Please pardon my noobishness!

[Aaron]

So if the two CPUs run at 2Ghz, does that mean my VPS slice is the equivalent of dual 125Mhz processors (2Ghz/8) and I can't utilise more than that? Or does it work more like virtuozzo in that if no-one else is using those two cores I get full access to them, but if there's heavy load involved I don't go *below* that?

If nobody else is using the cores, you get full access to them.

Either way, is my load average figure based on my 1/16th of a core, or does it reflect the total core - ie if everyone's accounts on the node were maxed out, would a load average of 0.0625 (1/16th of 1) mean my server was using all its CPU resources?

Load averages have to do with the number of processes waiting to be completed, and time it takes to complete them - so not so much to do with the speed of the CPUs. It does really tell you a percentage of a CPU/core used, but rather a look at the processes waiting for CPU time.

Hope this helps!

[jbngar]

Originally Posted by Aaron

If nobody else is using the cores, you get full access to them.

Thanks again Aaron

Part of the reason I moved from shared hosting to VPS (and a big factor in choosing Xen over Virtuozzo) was to eliminate the ability of others on the node to consume my resources.

I've heard something about 'cpu weighting' - does this mean that someone else on the node with a higher weighting (because they have a higher plan or whatever) gets priority on the processing queue? Or will I always have my 1/8th of a slice and if I need it I'll always be able to get it?

Putting it another way:
* is it physically possible for someone else on my node to affect my performance in ANY way?
* Is there a way to see how my VPS will perform should the other clients on the node be maxing out the CPU?

I'm just trying to avoid a situation like this:
* my VPS appears to be running SO fast and SO smooth because there's hardly anyone else on the node
* so I sell my services saying 'look how fast and smooth this is'!
* then as time goes on you guys add more VPSs to the node, which grab CPU time and lower performance
* I look like an idiot because my server gets slower and slower... :-b

I guess I don't want to be selling my 'Burst CPU' as 'Actual CPU'...

Thanks again for your time!






"each VPS operates as if it were a standalone server with dedicated RAM, Hard Drive Space and Processing power."

[jbngar]

Sorry Aaron - not sure if you saw these questions buried in the other rambling drivel...

* is it physically possible for someone else on my node to affect my performance in ANY way?
* Is there a way to see how my VPS will perform should the other clients on the node be maxing out the CPU?
* will I always have my 1/8th of a slice and if I need it I'll always be able to get it?

[Aaron]

Originally Posted by jbngar

Sorry Aaron - not sure if you saw these questions buried in the other rambling drivel...

* is it physically possible for someone else on my node to affect my performance in ANY way?

In terms of CPU, and RAM no. In some rare circumstances it is possible for other users to effect I/O (drives) on the server. However, this is quite rare and not a issue which usually comes up.

* Is there a way to see how my VPS will perform should the other clients on the node be maxing out the CPU?

At this point in time there is no way of testing this.

* will I always have my 1/8th of a slice and if I need it I'll always be able to get it?

You will always be able to access the CPU assigned to you, so no need to worry about that.