Planning Considerations for Fibre Channel Topologies
3
Fabric Initialization
McDATA Products in a SAN Environment - Planning Manual
3-30
When multiple directors or switches are connected, E_Port (ISL)
communication must be established between fabric elements and the
fabric must be initialized. During fabric initialization, the fabric
elements:
•
Establish the operating mode for connected E_Port pairs and
exchange link parameters (E_Port names, timeout values, class-
specific information, and flow control parameters).
•
Exchange fabric parameters, select a principal switch, and assign
Domain_IDs to all switches.
•
Employ a routing protocol to establish the shortest path through
the fabric and program route tables for each fabric element.
•
Exchange the active zone set to ensure uniform zoning is enforced
between all fabric elements.
Fabric initialization is not a serial process. The process executes
concurrently across all ISLs in the fabric, causing a massive flood of
Class F traffic that must processed to the embedded port of each
fabric element within a specified (fabric-wide) E_D_TOV. If the fabric
consists of a large number of elements (and therefore ISLs), Class F
traffic may not be processed within the E_D_TOV, resulting in error
recovery operations, timeouts, segmented links, or fabric failure.
Because of these problems, a fabric with a high ISL count is more
difficult to build. Problems associated with a large fabric are not
directly related to the large number of fabric elements but to the large
number of ISLs associated with the elements.
Installing high-port count directors (such as the Intrepid 10000
Director) as SAN building blocks provides a larger number of
non-blocking Fibre Channel ports. Large fabrics built around these
directors require fewer additional fabric elements (smaller directors
and fabric switches) and ISLs. The Intrepid 10000 Director also
supports high-bandwidth (10.2000 Gbps) ISLs that reduce the fabric's
total ISL count. In addition, fabrics and sub-fabrics can be merged or
maintained as separate entities through flexible partitioning
provided by the Intrepid 10000 Director.
Large fabrics benefit from deterministic non-blocking performance,
less ISL congestion, and better cable management - performance that
is not possible from a fabric constructed with smaller port-count
switches interconnected with multiple ISLs.