Technologies for the ProLiant ML570 G3 and ProLiant DL580 G3 Servers Technology Brief

errors and does not provide any capability of replacing failed or degraded DIMMs without shutting
down the server.
Online Spare memory reduces the likelihood of uncorrectable memory errors but does not protect
against uncorrectable memory errors. Like advanced ECC, Online Spare provides no capability of
replacing failed or degraded DIMMs without shutting down the server. The amount of memory
reserved for the Online Spare rank will vary from one system configuration to the next. The exact
percentage of memory available to the OS when using Online Spare mode depends on the number
and size of DIMMs populated per memory board.
Hot Plug Mirrored Memory provides protection against both correctable and uncorrectable memory
errors. Hot Plug Mirrored Memory also allows replacing failed or degraded DIMMs while the system
is operating. However, this mode uses half of the installed memory for redundancy.
Hot Plug RAID memory often provides the most economical and effective memory protection. Like Hot
Plug Mirrored Memory, it protects against correctable and uncorrectable memory errors, but does so
while allowing 75 percent of the installed memory to be available to the OS. Hot Plug RAID memory
also allows replacing failed or degraded DIMMs while the system is operating.
Table 1 summarizes the choices between levels of memory protection.
Table 1. Tradeoffs between hot add, hot replace, and amount of memory available for system use.
Memory Option Hot-add support Hot-replace support Memory utilization
Advanced ECC Yes No All
Online Spare No No Varies
Mirroring No Yes 50%
RAID No Yes 75%
Ensuring reliability within large-footprint memory designs
High-availability memory technologies are one way to assure the customer that the system will be
reliable as memory footprints grow larger. Another way to ensure reliability is to carefully choose
memory suppliers and components.
One of the challenges with increasing amounts of memory is the variation among memory suppliers.
Variations can occur at the design level as well as the production level. For example, although many
of the design parameters for a DIMM design are specified by JEDEC, the design of the I/O buffer
which drives and receives the signal is not specified. Therefore, each vendor designs their own
buffers, which determines the signal strength that is sent to or from memory. In addition, over the life
of a particular DIMM technology (for example, 512-MB DIMMs using DDR technology at 133 MHz),
the DIMM suppliers may change internal silicon designs to increase yield or fix minor bugs. Any of
these variations can cause potential problems in the platform if the variations are excessive.
HP employs multiple tactics to ensure quality memory components. First, HP uses industry-standard
components from top-level suppliers and makes sure that these components meet standard
specifications. In addition, HP works closely with chipset and component designers to ensure that
normal variations among components do not cause reflections or noise that can alter signal integrity.
HP engineers use utilities developed in-house to evaluate chipset and system noise margins before a
product is released to production. Finally, HP uses several levels of diagnostics during the server
manufacturing process: an in-circuit test to verify that components are placed correctly; a functional
board test to determine that the motherboard is functioning properly and can boot the operating
system; and a full system diagnostic to ensure that all the components in the system function properly.
15