The data plane software, which operates in active/active mode, manages flow processing and session state redundancy and processes transit traffic. All packets belonging to a particular session are processed on the same node to ensure that the same security treatment is applied to them. The system identifies the node on which a session is active and forwards its packets to that node for processing. (After a packet is processed, the Packet Forwarding Engine transmits the packet to the node on which its egress interface exists if that node is not the local one.)
To provide for session (or flow) redundancy, the data plane software synchronizes its state by sending special payload packets called runtime objects (RTOs) from one node to the other across the fabric data link. By transmitting information about a session between the nodes, RTOs ensure the consistency and stability of sessions if a failover were to occur, and thus they enable the system to continue to process traffic belonging to existing sessions. To ensure that session information is always synchronized between the two nodes, the data plane software gives RTOs transmission priority over transit traffic.
Before You Begin |
|---|
This topic includes:
The data plane software creates RTOs for UDP and TCP sessions and tracks state changes. It also synchronizes traffic for IPv4 passthrough protocols such as Generic Routing Encapsulation (GRE) and IPsec.
RTOs for synchronizing a session include:
The data link is referred to as the fabric interface. It is used by the cluster's Packet Forwarding Engines to transmit transit traffic and to synchronize data plane dynamic runtime state. When the system creates the fabric interface, the software assigns it an internally derived IP address to be used for packet transmission.
The fabric is a physical connection between two nodes of a cluster and is formed by connecting a pair of Ethernet interfaces back-to-back (one from each node).
Unlike for the control link, whose interfaces are determined by the system, you specify the physical interfaces to be used for the fabric data link in the configuration.
For SRX Series chassis clusters, the fabric link can be any pair of Ethernet interfaces spanning the cluster; for J Series chassis clusters, any pair of Gigabit Ethernet interfaces.
Table 110: Supported Fabric Interface Types for SRX Series Devices
The fabric data link does not support fragmentation. To accommodate this state, jumbo frame support is enabled by default on the link with an MTU size of 8980 bytes. To ensure that traffic that transits the data link does not exceed this size, we recommend that no other interface exceed the fabric data link's MTU size.
For JUNOS Software, flow processing occurs on a single node on which the session for that flow was established and is active. This approach ensures that the same security measures are applied to all packets belonging to a session.
A chassis cluster can receive traffic on an interface on one node and send it out an interface on the other node. (In active/active mode, the ingress interface for traffic might exist on one node and its egress interface on the other.)
This traversal is required in the following situations:
If the ingress and egress interfaces for a packet are on one node, but the packet must be processed on the other node because its session was established there, it must traverse the data link twice. This can be the case for some complex media sessions, such as voice-over-IP (VoIP) sessions.
 |
Note: Intrusion Detection and Prevention (IDP) services do not support failover. For this reason, IDP services will not be applied for sessions that were present prior to the failover. IDP services will be applied for new sessions created on the new primary node. |
The fabric data link is vital to the chassis cluster. If the link is unavailable, traffic forwarding and RTO synchronization are affected, which can result in loss of traffic and unpredictable system behavior.
To eliminate this possibility, JUNOS Software detects fabric faults and disables one node of the cluster. It determines that a fabric fault has occurred if a fabric probe is not received but the fabric interface is active.
To recover from this state, you must reboot the disabled node. When you reboot it, the node synchronizes its state and RTOs with the primary node.
|
Note: If you make any changes to the configuration while the secondary node is disabled, execute the commit command to synchronize the configuration after you reboot the node. If you did not make configuration changes, the configuration file remains synchronized with that of the primary node. |
