DIMM sparing provides a mechanism on SPARC M8 and SPARC M7 servers to unconfigure a failed DIMM with minimal performance loss. This feature allows deferred maintenance for DIMMs that are faulted and thus reduces the need for downtime. DIMM sparing is supported only on systems whose memory slots are fully-populated with DIMMs.
If a DIMM is diagnosed to be faulty during boot time or run time, the memory dynamically switches from 16-way to 15-way interleave by remapping all of the physical addresses to the remaining 15 DIMMs. To enable this remapping, the platform firmware must initially reserve space for the contents of one DIMM. Consequently, only 15 DIMMs worth of physical address space is made available to the system, even when 16 DIMMs are functioning.
DIMM sparing is enabled on individual CPU nodes (that is, CMIOUs) and can tolerate one DIMM failure per CMIOU. On a fully loaded SPARC M8-8 or SPARC M7-8 server with two PDomains, up to 4 DIMMs can be faulted per PDomain for a total of 8 faulted DIMMs on the server. For a SPARC M8-8 or SPARC M7-8 server with one PDomain, up to 8 DIMMs can be faulted, and for the SPARC M7-16 multihost server, up to 4 DIMMs can be faulted per DCU for a total of 16 faulted DIMMs on the server. These totals presume one DIMM per CMIOU is faulty. If a second DIMM fails in a single CMIOU, a service notification is issued and both DIMMs must then be replaced.
See DIMM Configuration in SPARC M8 and SPARC M7 Servers Service Manual for more information about memory configuration..