Hitachi Compute Blade 2000 User Manual page 1070

Memory space for a guest OS

Memory space available for a guest OS is calculated by subtracting the amount of
memory the logical firmware uses from the amount assigned to the LPAR. This
value is usually the same as the memory space displayed for the guest OS.
However, all assigned memory space may not be available depending on the guest
OS specifications or environment.
When the amount of memory for a guest OS is not enough
against the amount applications use, memory swap might
cause performance degradation. When assigning an amount
of memory, make sure to take the memory amount for
firmware as well as that used by the OS and applications into
account.
 NUMA (Non-Uniform Memory Access)
It is necessary to assign the processor and memory to the
LPARs while understanding the NUMA feature well when
using the NUMA setting.
LPAR manager recommends using the non-NUMA on all
server blades.
It is unnecessary to set NUMA when the NUMA features are
not required.
Outline
The server blade of the NUMA configuration has two or more nodes which consists of a
socket (group of processors) and memory connected to it. Nodes are connected by the
interconnection.
The memory in the same node for a processor is called local memory, and the memory
in the other node is called remote memory.
The processor and the local memory are connected directly and close. But the distance
to remote memory connected by an interconnection is physically further away
compared with local memory.
Each processor can use the memory in all nodes, but the access costs (a delay time) to
the memory are influenced by the physical distance.
The memory access cost to the local memory is smaller than the remote memory,
because the distance to the local memory is shorter than the remote memory. This
means that the local memory access performance is higher than the remote one.
1019