Step 13. One of the servers handling the call goes down

There are multiple roles and servers involved in serving the call: the signaling layer engages B2BUA and up to two instances of SG, the media layer engages MGC and the MG.

Falling MG. At the same time as MG goes down, all calls whose contexts are served on this MG stop transmitting traffic. Users stop hearing each other. The MGC monitors the crashes of its MG's MG servers and services and notifies the owners of the media contexts served on them, i.e., those roles that initiate requests to the MGC to create media contexts, in particular B2BUA. They, in turn, have the ability to transfer an active call to another MG in the same or a different group of MGC. The B2BUA performs this operation within 1 second. This process is called migration.

Falling _MGC_. Doesn’t lead to anything.

If MGC has a standby instance - control of the group, including active calls in it, goes to the standby, all MGs are also switched to it.

If there are no available standby instances of MGC in the group, then a reasonably quick restart will also result in no visible phenomena.

If the MGC collapses permanently in the absence of a backup instance, the current calls will continue, but will not be able to manage media traffic, in particular to activate melodies on hold, to detect DTMF signals arriving in media traffic (RFC-2833, Inband-audio). New calls, on the other hand, will be served on another group MGC, if any, or B2BUA will deny service at that site.

Falling B2BUA and SG. A drop of any of the SIP servers in the call control chain (SG-1, B2BUA, SG-2) leaves the call active and with continued exchange of media traffic, but without the ability to control the SIP call. Any command to end the call, put or take it off hold, transfer, form a three-party conference - leads to the termination of the call on the initiative of the party who tried to control it. The other party does not receive a disconnect, but hears silence.

At the same time, even restarting a fallen B2BUA will not restore call control, since the call context is not saved on hard drives, but is stored in RAM.

Restarting a fallen SG has a chance of restoring controllability to the conversation, but depends on what phase of the conversation SG fell and rose. Thus, if there was some exchange of control commands at the time of the fall, the chances are slim.