Reliability, Availability, And Serviceability (Ras) Features - IBM System x3850 X5 Implementation Manual

2.3.6 Reliability, availability, and serviceability (RAS) features

In addition to Hemisphere Mode, DIMM balance and memory size, memory performance is
also affected by the various memory reliability, availability, and serviceability (RAS) features
that can be enabled from the Unified Extensible Firmware Interface (UEFI) shell. These
settings can increase the reliability of the system; however, there are performance trade-offs
when these features are enabled.
The available memory RAS settings are normal, mirroring, and sparing. On the X5 platforms,
you can access these settings under the Memory option menu in System Settings.
This section is not meant to provide a comprehensive overview of the memory RAS features
that are available in the Xeon 7500 processor, but rather it provides a brief introduction to
each mode and its corresponding performance effects. For more information about memory
RAS features and platform-specific requirements, see the following sections:
System x3850 X5: 6.9, "UEFI settings" on page 259
System x3690 X5: 7.8, "UEFI settings" on page 337
BladeCenter HX5: 8.5, "UEFI settings" on page 396
The following sections provide a brief description of each memory RAS setting.
Memory mirroring
To further improve memory reliability and availability beyond error checking and correcting
(ECC) and Chipkill, the chip set can mirror memory data to two memory ports. To successfully
enable mirroring, you must have both memory cards per processor installed, and populate the
same amount of memory in both memory cards. Partial mirroring (mirroring of part but not all
of the installed memory) is not supported.
Memory mirroring, or
with a redundant copy of all code and data addressable in the configured memory map.
Memory mirroring works within the chip set by writing data to two memory ports on every
memory-write cycle. Two copies of the data are kept, similar to the way RAID-1 writes to disk.
Reads are interleaved between memory ports. The system automatically uses the most
reliable memory port as determined by error logging and monitoring.
If errors occur, only the alternate memory port is used, until bad memory is replaced.
Because a redundant copy is kept, mirroring results in having only half the installed memory
available to the operating system. FAMM does not support asymmetrical memory
configurations and requires that each port is populated in identical fashion. For example, you
must install 2 GB of identical memory equally and symmetrically across the two memory ports
to achieve 1 GB of mirrored memory.
FAMM enables other enhanced memory features, such as Unrecoverable Error (UE)
recovery, and is required for support of memory hot replace.
Memory mirroring is independent of the operating system.
For more information about system-specific memory mirroring installation options, see the
following sections:
x3850 X5: 3.8.4, "Memory mirroring" on page 87
x3690 X5: 4.8.6, "Memory mirroring" on page 141
BladeCenter HX5: 5.10.4, "Memory mirroring" on page 200
28
IBM eX5 Implementation Guide
full array memory mirroring redundancy,
(FAMM) provides the user