Specifications
DATA CENTER BEST PRACTICES
SAN Design and Best Practices 15 of 84
Brocade recommends core-edge or edge-core-edge as the primary SAN design methodology, or mesh topologies
used for small fabrics (under 2000 ports). As a SAN design best practice, edge switches should connect to at
least two core switches with trunks of at least two ISLs each. Each of those trunks should be attached to a
different blade/port group. In order to be completely redundant, there would be a completely mirrored second
fabric and devices need to be connected to both fabrics, utilizing MPIO.
Recommendations for switch ISL/UltraScale ICL connectivity are:
•There should be at least two core switches.
• Every edge switch should have at least two trunks to each core switch.
•Select small trunk groups (keep trunks to two ISLs) unless you anticipate very high trafc volumes. This
ensures that you can lose a trunk member without losing ISL connectivity.
•Place redundant links on separate blades.
•Trunks should be in a port group (ports within an ASIC boundary).
•Allow no more than 30m in cable difference for optimal performance for ISL trunks.
•Use the same cable length for all UltraScale ICL connections.
•Avoid using ISLs to the same domain if there are UltraScale ICL connections.
•Use the same type of optics on both sides of the trunks: Short Wavelength (SWL), Long Wavelength (LWL), or
Extended Long Wavelength (ELWL).
Device Placement
Device placement is a balance between trafc isolation, scalability, manageability and serviceability. With the
growth of virtualization and multinode clustering on the UNIX platform, frame congestion can become a serious
concern in the fabric if there are interoperability issues with the end devices.
Trafc Locality
Designing device connectivity depends a great deal on the expected data ow between devices. For simplicity,
communicating hosts and targets can be attached to the same switch.
fig09_SAN_Design
Data
Flow
Data
Flow
Minimal Data Flow
Figure 9. Hosts and targets attached to the same switch to maximize locality of data ow.










