Peripheral Gateway Fault Tolerance
Duplexed PGs are implemented to provide fault tolerance in Unified ICM software communication with peripherals. The duplexed PGs use a private network. The PG private network synchronizes certain processes within a duplexed PG pair. It also conducts "heartbeat detection", a process by which each PG sends a heartbeat packet every 100ms to keep track of the "health" of the other PG.
PGs use a combination of the hot standby and synchronization approaches to fault tolerance. In the hot standby approach, one set of processes is called the primary, and the other is called the backup. In this model, the primary process performs the work at hand while the backup process is idle. In the event of a primary process failure, the backup process is activated and takes over. In a duplexed PG system, the Peripheral Interface Manager (PIM) processes use the hot standby approach to fault tolerance.
In the synchronization approach, the critical process is duplicated on separate computers. There is no primary and backup. Both process sets run in a synchronized fashion, processing duplicate input and producing duplicate output. Each synchronized process is an equal peer. Cisco refers to these equal peers as a synchronized process pair. In a duplexed PG system, the Open Peripheral Controller (OPC) process operates as a synchronized process pair.
The following figure shows how to use hot standby and synchronization in a duplexed Peripheral Gateway.
The OPC processes communicate with each other through a private network connection and the Cisco Message Delivery Service (MDS). The MDS provides a synchronizer service which combines the input streams from the PIMs and PG Agents on both sides of the PG to ensure that both OPC processes see exactly the same input.
The OPC process activates PIMs and PG Agents on each side of the duplexed PG. The OPC process also supplies uniform message sets from various PG types to the Unified ICM central controller.
The PIMs manage the interface between different types of ACDs and the OPC. PIMs are duplicated on each side of the system and operate in hot standby mode. A PIM can be active on either side of the duplexed PG, but not on both sides at the same time. For example, in the preceding figure PIMs 1 and 2 are active on Side A; PIM 3 is active on Side B. The duplexed OPCs communicate with each other through the MDS to ensure that a PIM is active only on one side at a time.
The duplexed PG architecture protects against a failure on one side of the PG. For example, if an adapter card controlling access to an ACD fails, a hot standby PIM can use the alternate PIM activation path. As shown in the preceding figure, PIM3 was activated from Side B of the PG. This might be in response to an adapter failure between the Side A PIM3 and ACD3. In this type of failure scenario, the PG can maintain communication with the attached ACD.
Only one PG Agent actively communicates with a side of the central controller. When messages arrive at the central controller, they are delivered to both sides by the central controller Synchronizer process. The PG maintains idle communication paths to both sides of the central controller in case a switch-over to the other side of the central controller or PG is necessary.