Sharing Registrations Across Peers; Sharing Bandwidth Across Peers - Cisco TelePresence Administrator's Manual

Clustering and peers
Note: configuration data that is applied across all peers should not be modified on non-master peers. At best
it will result in the changes being overwritten from the master; at worst it will cause cluster replication to fail.

Sharing registrations across peers

When a cluster peer receives a search request (such as an LRQ, ARQ or an INVITE), it checks its own and
its peers' registration lists before responding. This allows all endpoints in the cluster to be treated as if they
were registered with a single VCS.
Peers are periodically queried to ensure they are still functioning. To prevent delays during call setup, any
nonfunctioning peers do not receive LRQs.
H.323 registrations
All the peers in a cluster share responsibility for their H.323 endpoint community. When an H.323 endpoint
registers with one peer, it receives a registration response which contains a list of alternate gatekeepers,
populated with a randomly ordered list of the IP addresses of all the other peers in that cluster.
If the endpoint loses contact with the initial peer, it will seek to register with one of the other peers. The
random ordering of the list of alternate peers ensures that endpoints that can only store a single alternate peer
will failover evenly across the cluster.
When using a cluster, you should change the registration Time to live on all peers in the cluster from the
default 30 minutes to just a few minutes. This setting determines how often endpoints are required to re-
register with their VCS, and reducing this to just a few minutes ensures that if one VCS becomes
unavailable, the endpoint will quickly failover to one of its peers. To change this setting, go to
> Protocols > H.323 > Gatekeeper > Time to
SIP registrations
The VCS supports multiple client-initiated connections (also referred to as "SIP Outbound") as outlined in
RFC 5626.
This allows SIP endpoints that support RFC 5626 to be simultaneously registered to multiple VCS cluster
peers. This provides extra resiliency: if the endpoint loses its connection to one cluster peer it will still be able
to receive calls via one of its other registration connections.
You can also use DNS round-robin techniques to implement a registration failover strategy. Some SIP UAs,
such as Jabber Video, can be configured with a SIP server address that is an FQDN. If the FQDN resolves
to a round-robin DNS record populated with the IP addresses of all the peers in the cluster, then this could
allow the endpoint to re-register with another peer if its connection to the original peer is lost.

Sharing bandwidth across peers

When clustering has been configured, all peers share the bandwidth available to the cluster.
Peers must be configured identically for all aspects of bandwidth control including subzones, links and
n
pipes.
Peers share their bandwidth usage information with all other peers in the cluster, so when one peer is
n
consuming part or all of the bandwidth available within or from a particular subzone, or on a particular pipe,
this bandwidth will not be available for other peers.
For general information on how the VCS manages bandwidth, see the
Cisco VCS Administrator Guide (X8.1.1)
live.
bandwidth control
Managing clusters and peers
Configuration
section.
Page 166 of 507