System information

24 Logical Partitions on System i5
Figure 1-22 Uncapped sharing mode for shared uncapped processor configuration
Adjust the uncapped weight for this logical partition. The uncapped weight will be used to
determine the portion of free processing units that will be distributed to this logical partition
among all shared uncapped logical partitions when two or more shared uncapped logical
partitions demand more processing units from the shared processors pool.
For example, if logical partition 1 has an uncapped weight of 90 and logical partition 2 has an
uncapped weight of 180, then logical partition 1 will receive 1/3 available processing units
from the shared processors pool when partition 1 and partition 2 demand more processing
units at the same time and there are available processing units in the shared processors pool.
HMC will automatically calculate the minimum, desired, and maximum number of Virtual
Processors for logical partition. You may change the virtual processors setting now or later
using Dynamic Logical Partitioning (DLPAR).
1.3.11 Considerations on using shared processors
Logical partitioning using shared processors allows greater flexibility and maximum processor
resources utilization. However, you must understand that using shared processors incurs
some risks and may impact logical partitions performance. Using shared processors from the
shared processors pool makes much more work for the Power Hypervisor.
Each logical partition with shared processors is only given some processing time in a few
milliseconds, determined from the processing units assigned to it — usually less than
processing times of one physical/dedicated processor. So, the logical partition does not
always have enough time to complete its job in one CPU cycle. This problem causes the job’s
data to be re-dispatched to memory and cache for the next CPU cycle. It increases the
chance that the data is no longer in the cache, or even in the memory any more, because
much data for many jobs is loaded into memory, then into cache. It must be reloaded from
memory, or at the worst case, from disk (DASD), which may take a longer time to complete.