McDATA StorageWorks 2/140 - Director Switch Planning Manual page 145

Large Fabric Problems
• Inability to consistently schedule maintenance downtime for each
SAN.
• Stranded resources. Unused ports in one SAN cannot be used by
applications in another (port limited) SAN, and expensive
resources (such as tape backup elements) cannot be easily shared
across SAN boundaries.
Fibre Channel fabrics configure and manage themselves, and require
operator intervention only upon failure. When a Fibre Channel
device connects to a director or fabric switch, the device receives a
unique 24-bit (three-byte) network address composed of domain,
area, and port bytes. This address is used for routing data through the
fabric. Domain identifiers (Domain_IDs) are typically reserved for
fabric elements (directors and switches), and range between 1 and 31
for McDATA products. A fabric element with a unique Domain_ID
then allocates the remaining two bytes (area and port) to provide
network addresses for devices.
The principal switch selection process ensures each fabric element
has a unique Domain_ID. This process determines which director or
fabric switch acts as the master in a newly-configured SAN and
allocates unique Domain_IDs to the remaining elements. Without this
process, two elements could have the same Domain_ID, resulting in
duplicate addresses and misrouting of data.
Principal switch selection is initiated by transmitting exchange fabric
parameter (EFP) frames to expansion ports (E_Ports) of all connected
switches until a user-defined fabric stability timeout value (F_S_TOV)
is reached. The F_S_TOV is proportional to the number of fabric
elements. If the fabric is large, the F_S_TOV is set higher and the
selection process takes longer.
During principal switch selection, a disruptive or non-disruptive
build fabric event occurs. A non-disruptive event preserves fabric
element Domain_IDs and does not require reassignment of network
address blocks. However, a disruptive event may cause elements to
acquire a different Domain_ID, which means each attached device
must re-login to the fabric to acquire a new network address.
Ironically, the self-configuring functionality provided by the Fibre
Channel architecture makes it more problematic to build large fabrics
or to extend fabrics over large distances. Fibre Channel fabrics tend to
become unstable long before reaching the maximum theoretical
switch count (239) because of ISL congestion, disruptive build fabric
events, or sporadic disruptions.
Implementing SAN Internetworking Solutions
Implementing SAN Internetworking Solutions
4
4-3