HP StorageWorks 8B - FC Entry Switch User Manual page 10

1
Introducing ISL Trunking
Routing without the ISL Trunking Feature
Prior to the implementation of the ISL Trunking feature, device-level load sharing was done
through Fibre Channel networks that created ISLs and operated with the Fibre Channel Shortest
Path First (FSPF) routing protocol. The protocol established and communicated the shortest paths
for data to be carried from source to destination. Load sharing multiple ISLs to connect two
switches was established by a round robin algorithm that assigned the next available ISL as part of
the link to connect a Host Bus Adapter (HBA) and storage. Depending on the fabric event
sequences, multiple HBAs shared an individual ISL. This type of load sharing sometimes caused
the fabric to have congested input/output traffic over the ISLs connecting two switches.
Although FSPF compliant switches ensure fixed routing paths, and guarantee that all frames are
delivered in order, congestion occurs if the aggregation of the stream exceeds the capacity of one of
the ISLs in the path.
For example, Figure 1-1 shows four untrunked ISLs with a maximum capacity of 2Gb/Sec each, for
a maximum throughput of 8Gb/Sec. Due to traffic that is not trunked, the throughput of the four
ISLs is: 1Gb/Sec + 1.5Gb/Sec +.5G + 1Gb/Sec + 1Gb/Sec, giving a total of only 5Gb/Sec, because
two 2Gb/Sec data streams are competing for the same path.
Figure 1-1 Routing without the ISL Trunking Feature
Routing with the ISL Trunking Feature
ISL Trunking ensures that all links are used efficiently, eliminating congestion on one link, while
distributing the load of the links. This feature is designed to significantly reduce traffic congestion.
As shown in Figure 1-2, four ISLs are combined into a single logical ISL with a total capacity of
8Gb/Sec.
The ISL Trunking feature distributes workload across all ISLs in a trunk. Each incoming frame is
sent across the first available ISL. As a result, transient workload peaks for one system or
application are much less likely to impact the performance of other parts of SAN fabric. Because
the full bandwidth of each physical link is available, bandwidth is efficiently allocated.
1-2 ISL Trunking User's Guide