A group of Logical partition shares the processing power from a pool of physical processors. While that can be a one line description to what shared processor partition are , there is a lot to it than just running by using the processing power from a shared processor pool.
In power4, there was no concept of shared processor partitions and required LPARS to have dedicated processors assigned to it. The disadvantage here was that a lot processing power may remain idle with one LPAR while other LPAR'S are craving for more cpu's.
Shared Processor Partitions deals with the scenario in a better way. It steals idle cpu cycles from one LPAR and gives it to another LPAR which needs more processing power. The assigning of processing power when multiple LPARS compete for processor is decided on weightage given to the LPAR which will be detailed a little later.
A deeper dive now
Partitions on Power5 can be broadly classified into 2 based on the allocation of processors.
1 Dedicated Processor Partitions
2 Shared Processor Partitions
Dedicated Processor Partitions are much like the power4 LPAR'S and has a minimum,desired and maximum value for processors assigned to it.
Shared Processor Partitions shares the processors from a pool of processors.
All the processors which are not assigned to dedicated partitions together form the shared processor pool.
More About Shared Processor Partitions.
As said earlier sharedprocessor partitions uses processing power from a shared pool of processors.
Shared Processor Partitions are further divided into two
1 Capped Partition
Capped partition can make use of processing units lesser than or equal to its Entitled capacity from the shared pool
2 Uncapped partition
Uncapped partition can use more processing power than its entitled capacity in case it needs.
In the event of two uncapped partition competing for cpu from shared pool the lpar with a higher weightage will get the cpu cycles.
The weightage of LPAR is a value between 0-254 and the default is 128 . A value of 0 as weightage for an uncapped LPAR is considered as a capped partition.
Entitled capacity is the amount of processing units that gets assigned to the LPAR when the profile is activated. It can be any value ranging from the minimum value specified in profile upto the Desired value. [Remember that the maximum value for processing units is just applicable in case of a DLPAR operation to add more processing units on the fly]
The simple idea of Shared procressing pools is for LPARs to use processing units as and when its required and then put it back to the shared pool when they don't need it , so that other LPARs which might need extra processing power can pick it up from there.
Another thing to take care of when planning a shared processor partition is the virtual processor values.
Power5 LPARS has minimum, desired and maximum values for virtual processors in addition to the processor settings. The cpu from the shared pool are dispatched to the virtual processor by the power Hypervisor.
The total processing units an LPAR holds can conceptually be viewed as equally divided among the virtual processors. A virtual processor cannot hold more than 1.0 processing unit, so the number of virtual processors should be chosen, so as to harvest the additional processing power the LPAR may need to acquire from the shared pool.
For Eg if you have an uncapped partition with 2 virtual processors and the entitled capacity of the LPAR is 1.0 processing units. As load increases the server will demand and acquire more processing power from the shared pool. The amount of processing units becomes 1.5 ... 1.75 and then 2.0 . Unfortunately the LPAR cannot take any more processing units from the shared pool even if the shared pool has free processing units ,because it has run out of virtual processor. If the number virtual processor we specified while creating the LPAR was 3, we could have taken one more processing unit out of the shared pool to meet the processing requirement of the LPAR.
To Sum up
The shared processor partition, provides a way by which Processing power can be efficiently used between a set of LPARS by using the processing power as and when required and not holding up udle cpu cycles or rather not wasting it. Additional control over the way the processing power will be allocated/distributed are provided by features such as capped and uncapped partition and also the weightage for the uncapped partition.
1 comment:
Sandz,good one...
Post a Comment