Specifications

DATA CENTER BEST PRACTICES
SAN Design and Best Practices 47 of 84
ARL is used to manage the bandwidth going into the cores based on the available WAN bandwidth. There may be
a single WAN connection or separate WAN connections between the sites. ARL is used to manage the BW from
the Brocade 7800s to the WAN connection. This example has a single WAN connection, although you could just
as well use more than one WAN connection. ARL is congured such that the oor value is set to the WAN BW ÷
the number of interfaces feeding the WAN; in this case, it is 4 (2 from each Brocade 7800). The ceiling value is
set to either the line rate of the GE interface or the available WAN BW. For example, if the WAN is an OC-12 (622
Mbps), the ceiling ARL value is set to 622 Mbps. The oor value is set to 155 Mbps. When all the interfaces
are up and running, they will run at 155 Mbps. In an extreme case in which three Ethernet interfaces are ofine,
the remaining FCIP Ethernet interface will run at 622 Mbps, continuing to utilize all the WAN BW and keeping the
RDR application satised.
All circuits have a metric of 0 or 1 associated with them, as shown in Figure 35. 0 is the preferred metric and
is used until all metric 0 circuits have gone ofine. After all circuits with metric 0 have gone ofine, then metric
1 circuits are used. This is most useful with ring topologies, in which one span of the ring is used with metric 0
circuits and, if the span fails, then the other span is used with metric 1 circuits. Both metric 0 and 1 circuits can
belong to the same FCIP Trunk (same VE_Port), which means that if the last metric 0 circuit fails and a metric 1
circuit takes over, no data in-ight is lost during the failover using LLL.
fig35_SAN_Design
FCIP
tunnel
FCIP
tunnel
Metric = 0Metric = 0
Metric = 1 Metric = 1
Figure 35. FCIP trunk circuits with metrics.
Brocade FCIP uses keepalives to determine circuit health. Keepalives are sent at the timer value divided by 5.
Each keepalive that arrives resets the count. If the counter reaches 5, the circuit is deemed ofine and goes
down. Massive IP network congestion and dropped packets can conceivably cause all ve keepalives to be lost
in transit, causing the circuit to go down. You do not want the keepalive timer to be set too short, because the
TCP sessions across the WAN have the ability to ride through very short outages and recover quickly. If the
timer is too short, this will not happen before going down, although a longer keepalive interval will take longer
to detect a bad circuit. FICON and FCP circuits have different default keepalive timer settings when they are
congured. An argument that indicates FICON must be added when you are conguring circuits that are used
for FICON. FICON has stricter timing than FCP and must have no more than a 1-second keepalive timer. FCP has
more exibility, and the default is 10 seconds; nevertheless, best practice is to also set the keepalive timer to 1
second unless the IP network tends to have congestion and deep buffers that inadvertently trigger FCIP
circuit drops.
Protocol Optimization
Brocade FCIP offers seven different protocol optimizations: FICON tape read/write, FICON XRC, FICON Teradata,
OSTP read/write, and FastWrite. With design practices, the most important concept to understand with protocol
optimization is that there must be a determinant path for both outbound and return communications of an
exchange. This means that all FC frames from an exchange must pass through the same two VE_Ports in both
directions. This is difcult to achieve if there is a fabric on each end and multiple VE_Ports across which those
fabrics can choose to send the data. In this case, there is no deterministic path.
Protocol optimization requires keeping the state of the protocol local to each end device. These states are kept
in what is called a state machine. Brocade extension equipment can maintain state for up to about 20,000