The mirroring function among servers utilizes iSCSI to read/write the data of a local disk on another node via the network.
The network which is used in the mirroring among servers need to be a private network. This network is usable also as other networks of the cluster system, such as the cluster interconnect, the administrative LAN, and the public LAN.
However, if you are concerned that the I/O processing to the network mirror volumes may affect on other communications or vice versa, you must have the network dedicated to the mirroring among servers.
The IPv4 address or the IPv6 address can be used.
For IPv6 address, a global unicast address and a unique local unicast address can be used.
A link-local unicast address cannot be used.
When configuring the network used for the mirroring among servers to be redundant:
Use Virtual NIC mode or NIC switching mode of Global Link Services (hereinafter GLS).
To use NIC switching mode of GLS, specify the physical IP address for the IP address used for the mirroring among servers.
To use NIC switching mode of GLS (Physical IP takeover), do not register the takeover IP address to the cluster application.
Do not change the monitoring time of GLS.
Redundancy mode | Shared with interconnect | Not shared with interconnect | ||
|---|---|---|---|---|
IPv4 | IPv6 | IPv4 | IPv6 | |
No redundancy | Y | Y | Y | Y |
Virtual NIC mode of GLS | N | N | Y | Y |
NIC switching mode of GLS | N | N | Y | N |
GLS: Global Link Services
When using IP aliasing, use IP address that is always available on the node.
To improve the system availability, avoid sharing the network used for the mirroring among servers with the cluster interconnect, the administrative LAN, and the public LAN. Configuring the network to be redundant is also recommended.
Do not allow any operations that disable the communication done by using the IP address set by the process described in "4.8 Disk Setting for Performing Mirroring among Servers."
(Operation examples)
Stopping NIC that is not redundant or the virtual interface of GLS by using the ifconfig command or the ip command of OS.
Stopping the virtual interface by using any GLS commands.
In the configuration of mirroring among servers, besides the disk of any other nodes, GDS also accesses the disk of own node by using iSCSI. If the operations described above are allowed on the operating node, the disk of both nodes cannot be accessed from the operating node. In this case, the disk of the operating node cannot be accessed by the standby node. This will cause the following errors.
The cluster application is switched over.
First one of the two netmirror slices that consist the netmirror volume where the I/O error is detected is detached from the netmirror volume, and the slice becomes INVALID.
If the slice on the node where the cluster application is switched becomes INVALID, the cluster application cannot be started on the switched node, and the service will stop.
Note
Sharing the network with the cluster interconnect or the administrative LAN
When the I/O speed (data traffic per second) from the application to the netmirror volume exceeds 50% of the bandwidth of the cluster interconnect, I/O to the netmirror volume may affect the heartbeat processing of the cluster system. In this case, do not share the network for the mirroring among servers with the network for the cluster interconnect.
Do not share the network for the mirroring among servers with the network for the administrative LAN as well when the I/O speed to the netmirror volume exceeds 50% of the bandwidth of the administrative LAN.
When sharing the network for the cluster interconnect with the network for the mirroring among servers, the CIP IP address cannot be used for the mirroring among servers. Assign another IP address to the network device specified in the cluster interconnect.
When sharing the network for the cluster interconnect with the network for the mirroring among servers, the network for the mirroring among servers cannot be configured to be redundant. In this case, the risk of operation being stopped will be increased as follows:
Due to a network error of the mirroring among servers, the disk on the standby node cannot be accessed from the operating node. As a result, the disk on the standby node is detached from the mirroring.
If the cluster application is switched at this time, the operation stops.
For details, see "7.16.11 Switchover of the Cluster Application when the Network is Disconnected."
Note
Sharing the network with the public LAN
When sharing the network between the public LAN and the mirroring among servers, if an error occurs in the network, an error occurs both in the communication of service application and in writing to the netmirror volume. In this case, the service will stop due to the following behaviors.
The slice on the disk of the standby node is detached because the disk on the standby node cannot be accessed from the operating node.
Triggered by the communication error of the service application, failover of the cluster application occurs.
On the new operating node, the slice on the disk of the own node is detached and cannot be used. The disk of the remote node cannot be accessed due to a network error. In this case, the service will stop.
To continue the service during the single failure of the network, configure the network to be redundant.
If the total data traffic per second may exceed the network bandwidth by sharing the network, configure the unshared network.
