It may be a n00b question, but i really want to know , How Entitled Capacity is less and Used CPU is more when there is no Free CPU is available in the managed system.
I have 5LPARs in a MS with Dual VIO.
Managed System CPU details
Available: 0.20
Assigned to partitions: 15.80
Configurable: 16.00
You need to also list how many Virtual CPUs an LPAR has to understand this.
If an LPAR doesn't use it's entitled capacity is donates it back to the pool and can be allocated to other LPARs based on need and weighting. so although it's entitlement may be quite low, the server finds and reallocates unused processor time, so the situation you describe is very common.
The best way - In my opinion - to see what the CPU on an LPAR is doing is to use the
vmstat 1
command and look at the ent% column as this shows how much CPU the LPAR is using relative to the entitled capacity.
Entitled Capacity means that the LPAR is guaranteed to have that amount of CPU available to it, and that is what the Assigned to Partitions number means, is the sum of all entitled capacities for LPARs in the server.
As ross.mather has said, if your entitled capacity is not used, it goes back into the resource pool. If your LPAR CPU entitlement is capped, it obviously won't share it. What's there to share if its capped?
I don't know where you are getting your information from about what is entitled and what's used etc.
Perhaps go on your HMC and show me the LPAR profile set up you have done.
Read on lssyscfg for more information.
First off, i 'd like to ask you to kindly refrain from using leet-speak around here. I am sure it will enhance your chances of getting an answer if you undergo the effort to write "you are" ("you're") instead of "ur". After all, we are professionals here and we talk that way. Thank you for your consideration.
Yes, your LPARs are capped, but that does not mean they can't give back to the pool the resources they do not need at the moment. "Capped" only means that they cannot allocate needed resources above a certain point (the "maximum" value). So, if your other LPARs are more or less idle and your LPAR Nr. 5 is very busy it could allocate additional processors released to the pool from the other LPARs. If they would start to need their resources themselves of course the LPAR 5 owuld have to give them back.
And this is what the concept i missed, I understand a bit what you mean(Red book confused me)
In NMON i could see the CPU usage was 120% , it means it taking some resources from shared pool.
But could you please brief me more like , how much limit an capped LPAR can share thier idle resources to shared pool and also how to find the available resources residing in shared pool.
For capacity planning NMON is not the ideal tool. I suggest giving LPAR2RRD a try (which is available freely too) or Ganglia (OS software and free).
OK: basically, for CPU and memory, there are 3 values: minimum, desired and maximum. "Minimum" is what the LPAR needs at all cost, without that it won't start. "Desired" is what it gets if there are no other constraints. "Maximum" is what it can allocate at the upper limit.
Now, "capped" means that such a maximum is in place. "uncapped" means there is no maximum value and the LPAR can allocate whatever it needs.
The art of capacity planning is to find sensible values for these attributes. If you set the minimum too low you burden the system with swapping around resources very oftenly, which will be bad for overall performance. If you set minimum values too high you lose flexibility and the system is less free in redirecting the resources where they are needed most.
Finding out what "sensible" values are is a complicated task which requires experience and knowledge about load characterstics of the applications involved. A general answer is hard to give.
I suggest studying the Redbooks nevertheless, they are very well written (even if their merit becomes notable only on second view). I suggest to read and study the characteristics of the POWER processor too, especially terms like SMT, cache usage virtual/logical/physical CPUs/cores should help you understand better the workings of your p5xx/p7xx system.
Everything that we've told you about donating unused cycles still holds true for Capped LPARS, the only real difference is that a Capped LPAR. Isn't allowed to exceed its Entitled Capacity.
NMON running an all the LPARs with the Consolidator spreadsheet for analysis is a good way to see how the CPU is being used across the system.
One more last clarification , post that this discussion can be closed.
&
So based on both statements, It means Capped LPAR can share their resources to Shared Pool but it cant get some CPU from Shared Pool when it was needed.
Whereas uncapped can share as well as get resources in pool?
Close, but not quite. In normal circumstances a capped LPAR (as well as an uncapped one) will get the "desired" value. If resources are available both will get more if they need it, but a capped LPAR will stop at "maximum", whereas an uncapped LPAR will get even more beyond this point.
