Dell EMC SmartFabric OS10 User Guide Release 10.5.1

Principal switch selection

A principal switch is a switch that assigns and maintains a unique domain ID across the fabric. During the principal switch selection, the switch with the highest priority becomes the principal switch. If two switches have the same priority, the switch with lower WWN becomes the principal switch. The valid range to set the priority is 1, and from 3 to 255. Priority 1 has the highest priority. A switch with priority 255 cannot become the principal switch. You can modify the switch priority only when the vfabric is in inactive state. You can activate the vfabric only by adding VLAN and fcmap configuration under the vfabric configuration view. Ensure to configure the switch priority before activating the vfabric.

An ISL that points towards the principal switch is called an upstream principal ISL, which is discovered during principal switch election. After completing the fabric reconfiguration, the local switch requests for domain ID allocation through an upstream principal ISL. An ISL that points away from the principal switch is called a downstream principal ISL. Link failure results in rebuilding the fabric. When the principal ISL fails and if no other path exists between the two affected switches, then the build fabric (BF) operation is triggered. If the backup link (nonprincipal ISL) is available, then the link failure recovery is triggered. Whenever the principal switch election is retriggered nondisruptively, the switches check if the previously assigned domain IDs match the newly elected principal switch. The switches remember the previously assigned domain IDs. If you have already configured the preferred domain IDs, during a switch reboot, the switches check for the preferred domain IDs. If you do not configure a preferred domain ID, the previously assigned domain ID is considered throughout the switch reload. When two different fabrics join and when both fabrics have the same domain ID configured, the reconfigure fabric (RCF) operation occurs as the domain ID overlaps. Any duplicate domain IDs assigned during a fabric merge are detected during the EFP exchange.

Fabric reconfigurations occur in two ways. They are nondisruptive reconfiguration (build fabric) and disruptive reconfiguration (reconfigure fabric). BF occurs when two configured fabrics merge and both the fabrics have nonoverlapping domain ID list. Reconfigure fabric occurs when both the fabrics have overlapping domain ID list, which can be detected through EFP exchange. Login request packets from F or VF ports are not served until the domain ID allocation is successfully completed.

The principal switch assigns a requested domain ID based on the availability of the ID. The FLOGI requests, received until the domain ID is assigned are silently dropped.

All the switches in the fabric must have identical E_D_TOV ranging between 1000 ms to 10,000 ms and R_A_TOV timer values ranging between 5000 ms and 10000 ms respectively. If there is a mismatch in the configuration value, that switch is isolated from the fabric.

Fabric shortest path first (FSPF)

FSPF is a link state path selection protocol. All the ISL links between FC switches are treated as point to point. FSPF tracks the state of the links on all switches in the fabric, and associates a cost with each link. FSPF computes paths from one switch to all other switches present in the fabric. FSPF computes the best path by adding the cost of the link that is traversed by the paths and by choosing the least path with the least cost to reach a particular domain ID from a switch. It also computes the best path between the switches that is based on the link cost. FSPF computes the shortest path from the local domain to all other domains available in the fabric and updates route details with the next hop to reach the shortest path. Hold-time interval must be elapsed between the two runs of shortest path first (SPF) run. This release of OS10 does not support incremental SPF run. E ports exchange FSPF hello packets periodically on the ISL to form and maintain neighbor adjacency. The FSPF link state updates (LSU) use this link adjacency to exchange the link state information of a switch, across the fabric. Each switch maintains a link state database that is based on this link state information. This link state database is used in the SPF (Dijkstra algorithm) to compute the shortest path to reach a switch in the fabric. The name server service uses these routes to synchronize the name server database across the fabric. Hence, FSPF helps in building the fabric connectivity. Configure the same hold-time value on all the switches to ensure a consistent route convergence, and to avoid intermittent forwarding loop. When you configure a shorter hold-time, the route update is faster. FSPF detects the link failures in the fabric and recomputes the next available shortest path to reach the destination domain. FSPF also updates the change in route such as addition of new link or removal of existing link and when the link goes up or down.

Distributed domain name server (DNS)

DNS is responsible for name server registration and management of Nx_Ports that are attached to the switch. The registered name entries are stored in the local database. Each switch exchanges its name server information with other switches in the fabric to maintain a synchronized and distributed name service.

Each switch in the fabric must distribute switch registered state change notification (SW_RSCNs) throughout the fabric whenever there is a change in its local database.

Topology changes trigger SW_RSCN, and NS queries. Each switch forwards the SW_RSCN packets on the FSPF computed path towards the other switches in the fabric.

System logs

The system log file contains the system logs for the following events:

• FC port operationally up
• Principal switch selection
• Domain ID assignment
• Port isolation
• Fabric or nondisruptive reconfiguration data
• Disruptive fabric reconstruction
• Error conditions when ELP or EFP exchange fails, and when the port goes into isolated state

Restrictions and limitations

This section lists the restrictions, and limitations of the multiswitch fabric feature.

• The multiswitch feature does not support Virtual E-ports (VE), BB_credit configuration, autoport mode, static FC route, zone merging, ESC exchange between switches, and switch port initialization.
• Only one vfabric is supported per switch in the multiswitch mode.
• Interoperability with other vendors, such as non-OS10 switches are not supported.
• Due to hardware limitations, multiswitch fabric feature does not support ECMP at the NPU level.
• Incremental update for shortest path route computation is not supported. The shortest path computation always runs for the entire fabric.
• ACL entries that are installed for control and data traffic use statically reserved CAM entries. Dynamic ACL space allocation is not supported.
• The switch supports zoning configurations like the F port mode. Configure the same zoning configurations on all switches in the fabric to avoid the Logical Unit Numbers (LUNs) being lost, during topology changes.

Multiswitch Fabric (E Port) Limitations

The following limitations are applicable in 10.5.1:

• The multiswitch feature does not support Virtual E ports (VE), BB_credit configuration, autoport mode, static FC route, zone merging, ESC exchange between switches, and switch port initialization.
• Only one vfabric is supported per switch in multiswitch mode.
• Interoperability with other vendors, such as non-OS10 switches are not supported.
• Due to hardware limitations, the multiswitch fabric feature does not support ECMP at the NPU level.
• Incremental update for shortest path route computation is not supported. The shortest path computation always runs for the entire fabric.
• ACL entries that are installed for control and data traffic use statically reserved CAM entries. Dynamic ACL space allocation is not supported.
• The switch supports zoning configurations like the F port mode. Configure the same zoning configurations on all switches in the fabric to avoid the Logical Unit Numbers (LUNs) being lost, during topology changes.