To eliminate single points of failure for networking, each
subnet accessed by a cluster node is required to have redundant
network interfaces. Redundant cables are also needed to protect
against cable failures. Each interface card is connected to a different
cable, and the cables themselves are joined by a connector such
as a hub or a bridge. In the case of FDDI networks, each interface
card is connected via a cable to a different concentrator. This
arrangement of physical cables connected to each other via a bridge
or concentrator or switch is known as a bridged net.
IP addresses can be associated with interfaces on a bridged
net. An interface that has an IP address associated with it is known
as a primary interface, and an interface that does
not have an IP address associated with it is known as a standby
interface.
Standby interfaces are those which are available for switching by
MC/LockManager if a failure occurs on the primary. When MC/LockManager
detects a primary interface failure, it will switch the IP addresses
and any associated connections from the failed interface card to
a healthy standby interface card.
A selection of network configurations is described further
inthe following sections. For a complete list of supported networks,
refer to the MC/LockManager Release Notes
for your version of the product.
Providing Redundant FDDI Connections |
 |
FDDI is a high speed fiber optic interconnect medium. If you
are using FDDI, you can create a redundant configuration by using
a star topology to connect all the nodes to two concentrators, which
are also connected to two routers, which communicate with the world
outside the cluster. In this case, you use two FDDI cards in each
node. The configuration is shown in Figure 2-2 “Redundant FDDI Configuration”. Note that the concentrators are connected
together using dual cables cross-connected Port A to Port B. The
routers must be configured to send all packets to both concentrators.
Using Dual Attach FDDI Stations |
|
Another way of obtaining redundant FDDI connections is to
configure dual attach stations on each node to create an FDDI ring,
shown in Figure 2-3 “Configuration with Dual
Attach FDDI Stations ”. An advantage
of this configuration is that only one slot is used in the system
card cage. In Figure 2-3 “Configuration with Dual
Attach FDDI Stations ”, note that
nodes 3 and 4 also use Ethernet to provide connectivity outside
the cluster.
The use of dual attach cards gives protection against failures
in both cables and connectors, but does not protect against card
failures. LAN card failure would result in a package switching to
another node.
Fibre Channel Switched Configurations |
|
Another type of high speed fiber optic interconnect is provided
with Fibre Channel, which is normally used with larger clusters
(4 to 8 nodes). MC/LockManager clusters configured on enterprise
parallel server (EPS) systems can use Fibre Channel switches to
provide redundant network paths among nodes. A four node example
with two switches is shown in Figure 2-4 “FibreChannel Interconnect
in a Four-Node Cluster ”.
For a two-switch topology, all primary paths must go into one switch
while all standbys must go to the second switch.
In the figure, the solid lines refer to the primary network
path among the nodes and the dotted lines refer to the standby path.
If there is a failure in one of the primary lines, a local switch
takes place on the node that has a failed primary line, and the
standby is used instead.
Printable version
