Locality-Optimized Resource Alignment for Superdome 2

12
is interleaved memory. The first three instances can be built easily, with each taking three quarters of
the resources on a blade. The fourth instance must be built from the remaining resources and is
therefore split across all three blades. The splintering of a virtual partition should be avoided when
possible, but sometimes it is unavoidable.
For this example, the processor and memory allocations would be as shown in the following diagram:
Figure 4. Four equal virtual partitions configured according to LORA rules
Whenever a virtual partition must be split across multiple localities, it is best to distribute all the
resources symmetrically across those localities. The first three virtual partitions, the ones that will use
three quarters of the resources on a blade, should be configured to use 3/8 of the processor cores
and 3/8 of the local memory on each of the two sockets on their blade. The unfortunate fourth
instance is totally splintered, by being forced to use the remaining resources: one core and the
remaining local memory on each of the six sockets. This fourth virtual partition will have poorer
performance than the other three. But its performance will still be somewhat better than if it were
using 100% interleaved memory, whereas the other three virtual partitions will perform significantly
better than if their partition were configured with 100% interleaved memory.
The commands that could be used to establish the processor and memory allocations are as follows:
vparcreate -N 2 -p q1 -a mem::6144
vparmodify -p 2:q1 -a socket:1/1/0:mem::21504 -a socket:1/1/1:mem::21504
vparmodify -p 2:q1 -a socket:1/1/0:cpu::3 -a socket:1/1/0:cpu::3
vparcreate -N 2 -p q2 -a mem::6144
vparmodify -p 2:q2 -a socket:1/3/0:mem::21504 -a socket:1/3/1:mem::21504
vparmodify -p 2:q2 -a socket:1/3/0:cpu::3 -a socket:1/3/0:cpu::3