Understanding Subnet Manager Routing - HP zl User Manual

• Local Identifiers (LIDs)
• Global Unique Identifier (GUIDs)
• Partition Key (P_Keys)
• Configures switch forwarding tables.
• Receives traps from Subnet Management Agents (SMAs).
• Sweeps the subnet, discovering topology changes and managing changes as nodes are added and
deleted.
Understanding the Subnet Management Agents (SMAs)
Subnet Management Agents (SMA) are part of the Subnet Manager. A SMA is provided with each node
and process packets from the Subnet Manager.
If an Subnet Manager is elected master, all of its components, including SA, are implicitly elected
master. If a Subnet Manager ceases to be master, all of its components cease responding to messages
from clients.
Subnet Manager Hot Standby
The master and slave subnet managers can be synchronized so the information in the master is carried
over to the slave in the event of a fail-over. Refer to
page 84 to configure SM hot standby.
The hot standby/database sync feature is used to synchronize the databases between subnet managers
running on separate chassis.
The Subnet Manager maintains a data base in the volatile memory of the master SM containing all
required information.
How is the synchronization done?
The database synchronization is accomplished in two stages:
• Cold Synchronization - This stage is initiated by the master SM when it is ready to start a
synchronization session with a standby SM. In this stage, all out of sync tables are copied from the
master SM to the standby SM.
• Transactional Synchronization - This stage is entered following successful completion of the cold
synchronization stage. In this stage, all database update transaction requests that are processed by
the master, are replicated to the standby.
What can cause a standby SM to become the master SM?
• A crash of the node running the current master SM.
• Partitioning of the subnet (e.g. due to link failure).
• Graceful shutdown of the master (e.g. for maintenance purposes).
What happens when a master subnet manager fails?
In the event of a failure:
• The standby subnet manager becomes the new master.
• The new master rebuilds the data base from information retrieved during the subnet discovery
phase.
• Existing LID assignments are retained, where possible.
• All ports are reset to force them to re-join multicast groups, re-advertise services, re-request event
forwarding, and re-establish connections.
• A "SlaveToMaster" event trap is generated to trigger any necessary processing by external
management applications.

Understanding Subnet Manager Routing

There are two different concepts associated with InfiniBand routing:
"Enable/Disable Database Synchronization" on
5