Administrator Guide

Fabric OS Administrator’s Guide 115
53-1002920-02
Chapter
4
Routing Traffic
In this chapter
Routing overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Inter-switch links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
Gateway links. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
Routing policies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
Route selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Frame order delivery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
Lossless Dynamic Load Sharing on ports . . . . . . . . . . . . . . . . . . . . . . . . . . 129
Frame Redirection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
Routing overview
Data moves through a fabric from switch to switch and from storage to server along one or more
paths that make up a route. Routing policies determine the path for each frame of data.
Before the fabric can begin routing traffic, it must discover the route a packet should take to reach
the intended destination. Route tables are lists that indicate the next hop to which packets are
directed to reach a destination. Route tables include network addresses, the next address in the
data path, and a cost to reach the destination network. There are two kinds of routing protocols on
intranet networks, distance vector and link state.
Distance vector is based on hop count. This is the number of switches that a frame passes
through to get from the source switch to the destination switch.
Link state is based on a metric value based on a cost. The cost could be based on bandwidth,
line speed, or round-trip time.
With the link state protocol, switches that discover a route identify the networks to which they are
attached, receiving an initial route table from the principal switch. After an initial message is sent
out, the switch only notifies the others when changes occur.
It is recommended that no more than seven hops occur between any two switches. This limit is not
required or enforced by Fabric Shortest Path First (FSPF). Its purpose is to ensure that a frame is
not delivered to a destination after the Resource Allocation TimeOut Value (R_A_TOV) has expired.
Fabric OS supports unicast Class 2 and Class 3 traffic, multicast, and broadcast traffic. Broadcast
and multicast are supported in Class 3 only.