Planning Superdome Configurations
Planning Superdome Configurations
Rules and Guidelines for Configuring a Complex
Chapter 166
Guidelines for High Availability
• Make sure that more than one cell in each partition has an I/O
chassis containing core I/O (partitions containing only one cell, or
connected to only one I/O chassis, are not recommended).
The cells attached to chassis containing core I/O should be the
partition’s lowest-numbered cells. Each I/O chassis that contains core
I/O should also contain cards for a boot device and networking, placed
as described under “Rules” on page 64, but only one chassis with core
I/O also has to have a removable-media controller card (this chassis
should normally be the one attached to the lowest-numbered cell).
This redundancy will allow you to reboot the partition and continue to
use it if any one cell connected to core I/O should fail in any respect
(e.g., a failure in the cell itself, the I/O chassis it's attached to, or the
core I/O card in the I/O chassis).
• As far as possible, keep all the I/O chassis for a single partition in a
single cabinet.
Add an expansion cabinet only when all the slots in all the I/O chassis
in the CPU cabinet(s) have been used up; and in a 64-way-capable
system, add a second expansion cabinet only when all the available
slots in the first expansion cabinet have been used up.
But see “Guidelines for Expandability” on page 68 for a possible
exception.
• If you are using an I/O expansion cabinet, make sure each partition
has at least one cell attached to core I/O inside the CPU cabinet.
• Put a core I/O card and a boot card in every I/O chassis.
• Within a partition, spread redundant cards across I/O chassis such
that if oneI/O chassis fails,you stillhave sufficient connections to run
the system.
• Attach critical disk devices to cards in more than one I/O chassis, and
hence to more than one cell.
Some disk devices (such as a disk array with two SCSI controllers)
can be connected to more than one card. Use such devices for root and
boot disks, and also for disks containing mission-critical data. This
will allow you to get to the data on the disk even if the cell to which it
is primarily connected is disabled (or one card or any single
connection fails).