FW V06.XX/HAFM SW V08.02.00 HP StorageWorks SAN High Availability Planning Guide (AA-RS2DD-TE, July 2004)

Planning Considerations for Fibre Channel Topologies
71SAN High Availability Planning Guide
Although connection of additional devices to a loop does not impact switch
bandwidth (1.0625 Gbps), it does adversely impact overall loop performance
because part of the bandwidth is dedicated to overhead instead of information
transmission.
Loop performance is a complex function of several factors, including the:
■ Loop round-trip time — The time required for a frame to travel completely
around a loop is a function of the propagation delay associated with each
H_Port and NL_Port and the time required to travel through the fiber-optic or
copper transmission medium. The addition of ports (through cascaded hubs),
devices, and cabling increases the round-trip time.
■ Number of loop tenancies — Each cycle of device arbitration, loop opening,
frame transmission, frame reception, and loop closing is called a loop
tenancy. A Fibre Channel operation, such as a small computer system
interface (SCSI) write command, may require several tenancies to complete.
Because significant overhead is associated with establishing and ending each
loop tenancy, an increase in tenancies decreases loop performance. To
decrease the number of loop tenancies, plan to limit the number of
arbitration-initiating devices installed on the loop.
■ Service rate — The loop service rate is the number of H_Port service
requests the arbitrated loop can process in a time period and is defined as one
divided by the average loop tenancy duration. Long-duration loop tenancies
decrease the loop service rate because select devices monopolize the loop.
High-bandwidth storage devices that can rapidly process input/output (I/O)
requests typically cause long-duration loop tenancies. Plan to limit the
number of such devices installed on the loop.
■ Loop utilization — Loop utilization is the term that describes how busy an
arbitrated loop is and is defined as the request rate divided by the service rate.
The request rate is the rate at which devices arbitrate for access to the loop
and is a function of applications using the loop, not the loop itself. As the
request rate increases due to additional devices being added to the loop, the
probability of contention for loop access and arbitration wait time increases.
In fact, as loop utilization increases, arbitration wait time increases
nonlinearly.
Shared mode operation does not fully use the switch’s capabilities and should be
used only when connecting legacy FC-AL devices that do not support switched
mode operation.