Information About Dynamic FCoE Using DFA
Dynamic FCoE Overview
Fibre Channel over Ethernet (FCoE) enables I/O consolidation. It permits both LAN and SAN traffic to coexist on the same switch and the same wire. This feature enables you to consolidate multiple separate networks into a single converged infrastructure.
Key values of I/O consolidation using traditional FCoE are as follows:
-
Elimination of separate network infrastructures for SAN and LAN traffic.
-
Reduction in hardware requirements, such as cabling and server interface cards (NICs and HBAs), and lowering capital expense.
-
Reduction in power and cooling requirements for fewer physical assets.
-
Increasing deployment agility for multiprotocol networks, which preserves long-term investments while preparing for future uncertainty in protocol needs.
By using DFA technology, you can take FCoE consolidation even further:
-
Create a logical, rather than physical, SAN A/B separation.
-
Efficiently load balance multiprotocol traffic within the data center.
-
Dynamically establish relationships between switches, reducing the possibility for human error during configurations.
-
Improved high availability percentages as the scale increases.
The DFA architecture provides an inherent multipath capability with redundancy to handle node failures. Fabric level redundancy is provided through a double fabric model (SAN A/SAN B). The separation of the two SANs is logically implemented as two different VSANs that map to two different VLANs (VLAN A and B). Fibre Channel traffic in SAN A becomes the FCoE traffic in VLAN A, the Fiber Channel traffic in SAN B becomes the FCoE traffic in VLAN B, and the LAN traffic is carried on one or more additional VLANs over the converged Ethernet infrastructure. In this logical environment, the VSAN A/VSAN B configuration protects against fabric-wide control plane failures.
The traditional method of hosts that connect to two separate SANs is still supported with the FCoE using DFA architecture. The host is connected to two different leaf nodes that host a disjointed set of VSANs. Beyond these leaf nodes, the fabric is converged on the same infrastructure, but the host continues to see two SAN fabrics.
The following figure shows a DFA topology with n spines (S) and m leafs (L). The m leafs communicate to each other through the n spines using DFA encapsulation.
FCoE creates an overlay of FCoE virtual links on top of the underlying Ethernet topology, irrespective of how that Ethernet topology is constructed and which protocol is used to compute the MAC address routes.
In a dynamic FCoE environment, the topology is developed using the leafs as FCoE Forwarder (FCF) switches that are forwarded through transparent spines.
FCoE hosts and FCoE storage devices are connected to a DFA topology through the leaf switches. In this configuration, only the leaf switches perform FCoE forwarding (only the leaf switches behave as FCFs); the spine switches just forward MAC-in-MAC encapsulated Ethernet frames that are based on the outer destination MAC address.
The following figure shows the logical FCoE overlay topology of VE_Port to VE_Port virtual links on a DFA topology.
Only the FCFs, that are implemented by the leaf switches are part of this overlay topology. This topology is seen by Fabric Shortest Path First (FSPF), for each FCoE VLAN. FSPF computes over which virtual link to forward an FCoE frame based on its DomainID (D_ID). A virtual link is uniquely identified by the pair of MAC addresses associated with the two VE_Ports logically connected by it. Identifying the virtual link is equivalent to identifying which MAC addresses to use for the FCoE encapsulation on the transport network.
Use Lm as the number of leafs that are feature enabled. The feature might not be enabled on all leafs. The FCoE mesh is basically the leafs where FCoE is enabled.
Dynamic Fabric Automation Overview
Dynamic Fabric Automation (DFA) is a data center network architecture designed to address the following challenges:
-
Large number of servers hosted with switches/routers requiring huge forwarding tables
-
VM mobility requirements
-
Control plane scaling requirements
-
Simplified management
DFA addresses these requirements by defining a management framework that simplifies provisioning of large networks, a distributed control plane to handle the scale requirements, and an enhanced forwarding paradigm to suppress flooding of control protocols to achieve faster convergence and perform efficient multicasting.
FCoE is a data center technology for storage networks. The FCoE functionality is primarily realized by using FCoE over FabricPath technology. Changes and enhancements have been made to the management aspects of FCoE (POAP/DCNM/Autoconfig) that make it work seamlessly on a DFA network.
It should also be noted that the SAN functionality is not any different from what is available with FCoE over FabricPath networks. All the fibre channel applications and protocols continue to function like classical VE.
A DFA network uses segmentation ID and related protocols in the fabric. Segementation ID based encapsulation is not used for FCoE traffic. FCoE traffic is carried across the DFA fabric using regular FabricPath encapsulation. Dynamic FCoE Using DFA enables the co-existence of DFA and FCoE over FabricPath on the same FCoE DFA leaf nodes. FCoE VLANs are legacy FabricPath VLANs.
Dynamic Fabric Automation and Fibre Channel Over Ethernet
The key value of Dynamic Fabric Automation (DFA) is the ability for simplified management through DFA. Hundreds of thousands of end points can be provisioned and managed in a simplified and streamlined manner. FCoE fabrics do not have similar scalability requirements. Only a few key configuration capabilities required by FCoE (interface, vlan, vsan, to name a few) are included in DCNM. For additional information about DCNM, see http://www.cisco.com/c/en/us/support/cloud-systems-management/prime-data-center-network-manager/tsd-products-support-series-home.html for details about the FCoE specific changes.
Realizing SAN A/B Separation
In the previous figure, the physical connectivity for the topology follows typical leaf/spine CLOS architectural best practices. Logically, SAN A and SAN B are isolated at the Top of Rack (ToR) switches physically. Once the traffic enters the DFA network, the storage traffic is logically separated (see the following figure) across the network where it is physically separated once more to the storage device edge.
Dynamic FCoE gains the additional redundancy that is inherent in the DFA network by using the increased spine connectivity. A larger network with a large number of spines means increased reliability and stability for the storage network. This is achieved while retaining the best practices requirements for storage environments.
Load-Balancing FCoE Traffic on a Dynamic VFC
DFA provides redundant paths between a source and destination. Because FCoE traffic traverses the DFA network with one or more FCoE and non-FCoE nodes (spines, leafs), you must ensure in-order delivery through proper port-channel hashing across the redundant paths. All DFA nodes, both leaf and spine, have port-channel hashing enabled that includes the exchange ID. Traffic from a single flow always traverses through only one set of nodes through the network to maintain in-order delivery.
Supported Dynamic FCoE Using DFA Topologies
The supported topologies for Dynamic FCoE Using DFA are as follows:
-
FCoE devices that are directly connected to an FCF leaf
-
Traditional FCoE VE_Port connectivity to an FCF leaf
-
Legacy FC fabric connected to an FCF leaf
-
NPV and FCoE NPV devices that are connected to an FCF leaf
-
Native FC devices that are directly connected to an FCF leaf
Note |
Only classical FCoE is supported on an NPV device that acts as a leaf switch. Dynamic FCoE using DFA is not supported on an NPV leaf switch. |