Step 13. Failure of one of the servers serving a call

Several roles and servers participate in call servicing: the signaling layer uses B2BUA and up to two SG instances; the media layer uses MGC and MG.

Case: user A is in a call with B. During their conversation, the media server serving the call fails. The conversation does not stop; the subscribers hear each other (and in a video call, see each other).

MG failure. Simultaneously with the MG failure, traffic stops in all calls whose contexts are served by that MG. Users stop hearing each other. MGC monitors failures of MG servers and services in its group and notifies the owners of the media contexts served on them, that is, the roles that initiate requests to MGC to create media contexts, in particular B2BUA. They, in turn, can migrate the active call to another MG in the same or another MGC group. B2BUA performs this operation within 1 second. This process is called migration.

migration

MGC failure. It does not lead to anything.

If MGC has a standby instance, control of the group, including active calls in it, passes to the standby one, and all MGs also switch to it.

If there are no available standby MGC instances in the group, then with a sufficiently fast restart no visible effects will occur either.

If, in the absence of a standby MGC instance, MGC goes down for a long time, current calls continue but cannot control media traffic, in particular activate hold music, detect DTMF signals arriving in the media traffic (RFC-2833, inband audio). New calls will be served in another MGC group, if one exists, or B2BUA will reject service on this site.

B2BUA and SG failure. Failure of any SIP server in the call control chain (SG-1, B2BUA, SG-2) leaves the call active with ongoing media traffic exchange but without SIP call control. Any command to end the call, hold or unhold, transfer, or create a three-party conference causes the call to terminate on the initiative of the side that attempted the control. The other side does not receive a hangup but hears silence.

Even a restart of the failed B2BUA will not restore call control, because the call context is not persisted to hard drives but stored in RAM.

Restarting a failed SG carries a chance of restoring conversation control, but it depends on the phase of the conversation in which SG failed and restarted. For example, if some control command exchange was in progress at the time of failure, the chances are small.

Table 1. Terms used
Term Definition

Migration

!