IBM x3850 X5, Руководство по внедрению

Страница: 160 / 580

140
IBM eX5 Implementation Guide
access only to remote or “far” memory. So, threads executing on Processor 1 have a longer
latency to access memory as compared to threads on Processor 0. This situation is due to
the latency penalty incurred to traverse the QPI links to access the data on the other
processor’s memory controller. The bandwidth to remote memory is also limited by the
capability of the QPI links. The latency to access remote memory is more than 50% higher
than local memory access.
For these reasons, we advise that you populate all processors with memory, remembering the
requirements that are necessary to ensure optimal interleaving and Hemisphere Mode.
Figure 4-17 Memory latency when not spreading DIMMs across both processors
4.8.5 Mixing DIMMs and the performance effect
Using DIMMs of various capacities is supported for several reasons:
򐂰 Not all applications require the full memory capacity that a homogenous memory
population provides.
򐂰 Cost-saving requirements might dictate using a lower memory capacity for part of the
platform’s DIMMs.
Figure 4-18 on page 141 illustrates the relative performance of three mixed memory
configurations as compared to a baseline of a fully populated memory configuration. While
these configurations use 4 GB (4R x8) and 2 GB (2R x8) DIMMs as specified, similar trends
to this data are expected when using other mixed DIMM capacities. In all cases, memory is
populated in minimum groups of four, as specified in the following configurations, to ensure
that Hemisphere Mode is maintained.
Figure 4-18 on page 141 shows the following configurations:
򐂰 Configuration A: Full population of equivalent capacity DIMMs (2 GB). This configuration
represents an optimally balanced configuration.
򐂰 Configuration B: Each memory channel is balanced with the same memory capacity, but
half of the DIMMs are of one capacity (4 GB) and half are of another capacity (2 GB).
򐂰 Configuration C: Eight DIMMs of one capacity (4 GB) are populated across the eight
memory channels, and four additional DIMMs of another capacity (2 GB) are installed one
per memory buffer, so that Hemisphere Mode is maintained.
򐂰 Configuration D: Four DIMMs of one capacity (4 GB) are populated across four memory
channels, and four DIMMs of another capacity (2 GB) are populated on the other four
QPI links
Intel Xeon 7500
Processor 0
DIMM
DIMM
DIMM
DIMM
DIMM
DIMM
DIMM
DIMM
DIMM
DIMM
DIMM
DIMM
DIMM
DIMM
DIMM
DIMM
DIMM
DIMM
DIMM DIMM
Buffer Buffer Buffer Buffer
Memory
controller
Memory
controller
Intel Xeon 7500
Processor 1
DIMM
DIMM
DIMM DIMM
Buffer Buffer Buffer Buffer
Memory
controller
Memory
controller
LOCAL
REMOTE