Fault Avalanche Control

Feature Overview

Packets that cause a direct SBC fault can lead to a catastrophic failure of an SBC service, which is known as a packet-stimulated fault avalanche. These packets appear for various reasons, such as: the SBC adds a new Session Initiation Protocol (SIP) endpoint, upgrades or replaces a peering endpoint or gateway (GW), changes a configuration on a peer, or introduces a new call scenario. The SBC does not currently check for double faults, which is when the SBC has a failover and then another failover. Double faults cause call loss.

The goal of Avalanche Fault Detection and Control is not to prevent individual crashes, but to detect and control continuous "avalanche" faults that can lead to complete service outages. This feature uses the information from the existing faults to attempt to prevent future faults.

The fault avalanche feature tracks potentially problematic values of key types in SIP packets (see Tracking for more information). To track these values the fault avalanche feature extracts and saves values from the following fields of the SIP packet that causes a crash:

• Call ID
• Called Party Address
• Calling Party Address (From or P-Asserted-Identity header)
• Source IP of the Packet

Each key type associates with a threshold value. The threshold value indicates the maximum amount of crashes allowed for a particular value of the key type before the SBC blocks that key value (see Blocking for more information). The SBC defines the threshold values so that they ensure the threshold for more specific blocks trigger before less specific blocks. The following table shows the default key type thresholds.

Table 1: Default Key Type Thresholds

The SBC determines the calling party for a SIP packet in the following order:

1. If present, the P-Asserted-Identity header user part or telephone number
2. If present, the P-Preferred-Identity header user part or telephone number
3. If not anonymous, the From header user part or telephone number

The SBC obtains the called party for a SIP packet from either a Request URI of a SIP request, or the To header URL of a SIP response.

The Layer 3 source IP address of the packet determines the peer IP address.

Tracking

Before processing every SIP message, the SBC records key elements in a special location. When a crash occurs, the SBC will take these key elements and use them to create the Fault record. On every crash, the SBC creates a fault record that contains the values of the key types (see the Default Key Type Thresholds table) of the SIP packet that causes the failure. This fault record, which survives a reboot, locally saves and broadcasts to all other SBC instances in the cluster. The SBC can discard the packet to avoid another fault if there is a match.

Every SBC maintains a table of the local crash counts by key type and key value with the data from the fault records. The default aging period for fault records is 30 minutes, so the SBC maintains records only within the last 30 minutes.

A fault record sends to the standby SBC instance within two seconds of the fault. For the SBC SWe Cloud, a fault record broadcasts throughout the SBC cluster within ten seconds of the fault.

The SBC removes a fault record that the SBC does not consider for tracking or protection functionality within five minutes of expiration. The expiration interval is a global configurable (faultAvalancheControl  faultRecAgeingTimeOut).

The SBC discards the fault record that it receives from another SBC in the cluster if the fault record is a duplicate of an existing fault record. Otherwise, the SBC treats the fault record as if it locally generated.

Blocking

The system maintains a (logical) block table with an entry for every key type and key value combination where the number of all unexpired fault records exceeds the threshold for that key type.

The block entries update within two seconds of when the SBC receives and, or, creates a new fault record. The block entries update within five minutes of a fault record expiration.

When the SBC receives a SIP Request or Response message from another SBC(s), the system drops the packet if any key type or key value matches a blocking entry.

The following figure illustrates the input processing of received SIP messages for the fault avalanche feature.

 Figure 1: Input Processing of Received SIP Messages 

Bootstrap

When an SBC SWe Cloud instantiates, it obtains all unexpired fault records. Subsequent processing is the same whether the SBC SWe Cloud creates the fault records locally, or receives the fault records from another SBC in a non-bootstrap scenario.

After an SBC SWe Cloud reboots and initializes, it obtains all unexpired fault records. Subsequent processing is the same whether the SBC locally creates the fault records or receives the fault records from another SBC in a non-reboot scenario.

When an SBC reboots, it uses locally persisted fault records to initialize. Subsequent processing is the same whether the the SBC SWe Cloud creates the fault records locally, or receives the fault records from another SBC in a non-reboot scenario.

An SBC uses the present fault records when that SBC activates from standby to active.

Provisioning and Monitoring