memory and cpu unit

[suppliers]

Im curious what is the unit of memory usage and CPU usage in /proc/lve/list file?
And how I can translate these numbers to Megabyte and Mega Hz?

Any help will be much appreciated.

[iseletsk]

For CPU is Hz (in CL6/lve0.9 we plan to switch to nano-seconds)

For memory: 4k pages

[clm]

Have you moved to nanoseconds ?

If so then does this mean we can set less than 1% for CPU limits ?

Our next server will be running CloudLinux 6 and the configuration we are currently looking at has got 32 threads of 3.1 GHz, this makes 99.2 Ghz which means we have got to choose between 1.98 and 2.97 Ghz to remplace our current 2GHz limit per account.

If we could set a decimal after the percentage it would be great ! (or at least half a percent…)

2.1% would be 2.08 GHz which would be perfect for our 2GHz accounts !

[iseletsk]

Richard, yes we moved to nanoseconds in /proc/lve/list for lve1.x kernels.
Yet, the way to setup CPU limits is still the same.
No way to represent decimal points yet

[clm]

Thanks for the info,

Do you plan to try and integrate decimal points this year ? Do you have even an aproximate idea of if and when this might be integrated into the current system ?

[iseletsk]

Richard,

We plan to let admin setup % as relative to individual core (which should be enough I believe) -- but we will not support decimal points.
I think this is something that might happen before the end of this year. This is mostly for sanity in setting limits, as so far you are the only one who requested better granularity.

[clm]

A percetage per core would be great, it would actually be better then 0.1 decimal points…

With this setting I could get 2.02GHz which is closer than 2.08 GHz

I presume that we would be able to set 1 core + 65% of a core… Would this be represented as 165% ?

[iseletsk]

No, you would just set: 165% - which would mean 1 core + 65% of another core.

[clm]

Thanks that will be perfect

[clm]

Hello,

Do you have an update on this CPU limit per core instead of per server ?

Our latest server has got 32 * 3.1Ghz threads and the charts in the statistics don show much detail when only 2 or 3% CPU is allowed and 1% equals 0.99 GHz of virtual CPU (threaded).

[iseletsk]

This is something I plan for Q1 2013. It is a small change, but given that so many customers would need to change their settings -- we want to plan it carefully.

[clm]

Great, thank you

[clm]

I the mean time, is there something that could be done for the details in the statistics ?

With 3% allowed CPU

I see only the following percentages in the CPU usage :

0% => 0% total CPU usage
33% => 1% total CPU usage
66% => 2% total CPU usage
100% => 3% total CPU usage

I would like to have at least 10 different values so the users can get a more precise idea of what they are using.

Would it be possible to process the stats on less than 1% total CPU usage increments before implementing the ability to set more precise settings ?

Does this impact on the actuel limitations ? or just on the stats generation ?
Thanks

[iseletsk]

The exact CPU accounting info is in /proc/lve/list

It is in nanoseconds (for CL6), and in ticks for CL5
You can try accumulating it/using it instead, but it is quite a lot of work.

[clm]

Thanks for the info.

Can we hope to have this used in the stats in the future ?

I dont need the increments used in nano seconds but would like to have no more than 5% increments…

When you allow 25%CPU the stats look fine, even with 10% CPU but if you allow 1% CPU (about 1GHz in my case) the stats would always be 0% or 100% with nothing inbetween…)

Thanks