AIX CPU
Micropartitioning –Better utilization of CPU resource
Entitled Capacity—Guaranteed amount of Proccessing Units
Processing Units: Each partition is configured with a percentage of execution dispatch time for each 10 msec timeslice (dispatch window)

Say EC=2.5
PU    Time with in dispatch window of 10 msec

1. 1
2. 2
3. 3
4. 4
5. 5
6. 6
7. 7
8. 8
9. 9

1. 10
2. 20

2.5    25 msecs


So 25 milliseconds will be divided between the 3 processors..8.x across each


But, the maximum PU's that's available for LPAR-1 is 3.0—30 milliseconds. This can happen only if the LPAR is uncapped.
Conferred calls: Virtual processor cycles of one processor is donated to another if it's waiting on the other's input
Simultaneous Multi-threading
    Enabled by default
    Can be disabled for scientific applications—number crunching
Each VP's in POWER 6 will use 2 logical CPU threads
Each VP's in POWER 7 will use 4 logical CPU threads
Maximum processor pools supported in POWER6 --64
Can be controlled by smtctl
Commands
Sar
Lparstat
Mpstat
Lsattr –E1 proc0
Power Hypersvisor schedules all the micropartitioning
Schedo command can be used for tuning hypervisor scheduler
VP_needed= physc+Vpm_xvcpus
Context switching:Too many processor can lead to lot of time spent on context switching
Virtual Processor folding


Dedicated Processor
Processor and memory affinity for best performance
Unused capacity lost
Less granular units of allocation
Can work in donating mode



Capacity entitlementàEntitled Pool capacityàMax Capacity
Virtual Processor Resolution Level
Level-0 resolution
If cycles required for an LPAR can be resolved within the same pool
Level-1 resolution
Available cycles available over and above the cycle requirement can be consumed by a different pool
Virtual Ethernet Adapters (Firmware)
LPARs communicate between each other using virtual ethernet adapters
(tagging of datagrams)
Virtual I/O Servers
When LPARS wants to communicate across the network
VLAN—priorities can be established
SEA-Shared Ethernet Adapters
SCSI—Storage
Threading/non-threading
Here threads are used for transferring datagrams
Virtual scsi and virtual Ethernet can communicate each other using threading
Threading hampers performance if used when SCSI is not there
When threading is enabled, a buffer is created that the kernel thread
When threading is not enabled, an interrupt is used for communication
Integrated Virtual Ethernet/Host Ethernet Adapter



Multi-core scaling –MCS


Virtual SCSI
Queue depth value should have the same as VIO server
N_Port_ID_Virtualization
LUN??


Partition mobility
Active memory sharing



Active memory expansion—works based on the memory compression set by the expansion factor. Can be monitored using lparstat Target memory expansion/target memory expansion size parameters
%xcpu: Percentage of CPU utilization used for AME compression and decompression activity
dxm: Memory deficit in MB
Command
Amepat
SMT
A logical processor has many registers namely, floating point, integer, load and save – they are not used at the same time, so using SMT enables the usage of the registers at the same time. If 2 registers have to be used at the same time then there is collision and there is a wait.


Process/Thread









WPARS
Application WPAR
System WPAR