IBM RS/6000 SP, Руководство по определению проблем

Страница: 129 / 316

This soft copy for use by IBM employees only.
Notice that in the SP Switch, the Estart_sw calculates the timeout based on the
number of switches:
LIMIT = 180 + ((NUM_SWITCH - 2) * 60) where NUM_SWITCH > 2
The following steps are referred to as Switch Initialization:
1. An Estart is issued for a system partition from the Control Workstation,
through either SMIT or the command line, and this is relayed to the primary
node of the partition as an
Estart_sw.
2. The primary node consults the SDR for the current topology file and the
current fenced nodes.
3. The primary node distributes the topology file as appropriate:
• The primary node distributes the topology file to the bootserver nodes of
its partition.
• The primary node directs each bootserver to distribute the topology file
to its client nodes.
• Each bootserver distributes the topology file, tests for success, and
returns resulting status to the primary.
• The primary node logs and evaluates failure data.
4. The primary runs the Worm code to verify the partition
′
s fabric pool and
nodes:
•
Each chip is sent a Read Status service package to check cabling.
• Each chip is sent an Initialization service package to set the chip ID and
specify routes to the primary.
• Each non-primary unfenced node is sent an Initialization service package
to set its personality and specify the topology file for this partition.
• Each non-primary unfenced node is sent a Read Status node service
package to verify the switch address, personality, and readiness.
• The primary node updates its device database in accordance with the
findings.
5. The primary node sets the Time-of-Day (TOD) across the fabric pool:
•
Traffic is quiesced, if necessary.
•
The TOD is propagated.
•
Traffic is resumed, if necessary.
6. The primary node downloads its route table to its TB3 adapter
7. The primary node distributes device database deltas to the nodes over the
switch through the
Device Database service package
8. On each non-primary, unfenced node:
•
The local device database is updated with the deltas.
• Routes are computed.
•
The routes are loaded to TB3 SRAM.
9. On each node, the communications protocols and Load Leveler code, as
appropriate, are notified that the switch is available.
Chapter 4. The Switch
109