Predefined policies—You can apply a predefined QoS policy: default-fcoe-in-policy. User-defined policy—You can create a QoS policy that conforms to one of the system-defined policies. For all the other Cisco Nexus Platform switches, this step is optional.
The interface ID range is Example of explicit vFC bound to interface port-channel:. Example of explicit vFC bound to break-out port:. Example of NP interface configuration using explicit vFC:. Example of F interface configuration using explicit vFC:. It breaks the FCoE path completely. Example of implicit vFC bound to interface port-channel:. Example of implicit vFC bound to break-out port:. Example of NP interface configuration using implicit vFC:. Example of F interface configuration using implicit vFC:.
Optional switchto vdc vdc-name. This step is required only when a Cisco Nexus Series switch is used as the core switch. Optional feature fport-channel-trunk. Configures the interface as a Layer 2 interface and deletes any configuration specific to Layer 3 on this interface. Configures the interface as a Layer 3 interface and deletes any configuration specific to Layer 2 on this interface.
The bind interface ethernet command is not required for an implicit vFC configuration. Exits the interface configuration mode. To put an interface that is in Layer 3 mode into Layer 2 mode for Layer 2 configuration, use the switchport command in interface configuration mode. To put an interface into Layer 3 mode, use the no form of this command. This step is optional. You can enable or disable a pause frame timeout value on a port.
The system periodically checks the ports for a pause condition and enables a pause frame timeout on a port if it is in a continuous pause condition for a configured period of time. This situation results in all frames that come to that port getting dropped in the egress. This function empties the buffer space in the ISL link and helps to reduce the fabric slowdown and the congestion on the other unrelated flows using the same link.
The QoS policy for uplink ports and server or host interfaces should be same for the slow drain feature to be operational. When a pause condition is cleared on a port or when a port flaps, the system disables the pause frame timeout on that particular port. The pause frame timeout is disabled by default.
We recommend that you retain the default configuration for the ISLs and configure a value that does not exceed the default value for the edge ports. For a faster recovery from the slow drain device behavior, you should configure a pause frame timeout value because it drops all the frames in the edge port that face the slow drain whether the frame is in the switch for a congested timeout or not.
This process instantly clears the congestion in the ISL. Use the no system default interface pause timeout milliseconds mode edge command to disable the pause frame timeout value on the edge ports.
The default pause timeout value is milliseconds. Configures a new pause frame timeout value in milliseconds and the port mode for the device. Timeout value is specified in multiples of range is The system default interface pause timeout milliseconds mode core command is not supported.
Configures the default pause frame timeout value in milliseconds and the port mode for the device. Only the system default interface pause milliseconds mode edge command is supported.
The system default interface pause milliseconds mode core command is not supported. Disables the pause frame timeout for the device. Disables the default pause frame timeout for the device. This example shows how to configure a pause frame timeout value:. This example shows how to display pause frame timeout information:. This example shows syslog messages that are displayed when a pause frame timeout occurs:.
To display NPV configuration information, perform one of the following:. Set switch side physical interface to trunk mode. Set server side physical interface to trunk mode. Skip to content Skip to search Skip to footer. Book Contents Book Contents. Find Matches in This Book. PDF - Complete Book 2.
Updated: December 22, Example: switch config npv auto-load-balance disruptive Traffic mapping You can specify the NP uplinks that a server interface can use to connect to core switches. Save the configuration.
FIP is thus solving this chicken-and-egg problem by handling the whole fabric login process. This additional configuration ensures the uniqueness of the produced FPMA in the whole network. In native FC, both ends of a physical links are implementing an FC stack. If the link fails, an indication is provided to FC. FIP allows for detecting these kinds of failures by providing a basic keepalive mechanism.
However, in the case of FCoE to the access, where FCoE is only running over a single point-to-point link, this capability has little use. It is a solution with no gateway: just like FCIP, it encapsulates FC traffic over a different medium while maintaining all the characteristics of native FC.
FCoE assumes a lossless Ethernet network. This fundamental function is provided with the help of the IEEE standards introduced in the next section. This means that the converged network must be able to simulate independent links for different classes of traffic.
For example, you might want to reserve a certain bandwidth for FC traffic so that a burst of IP packets cannot cause a loss of FC frames. ETS defines priority groups to which various frames can be assigned based on their priority and how to allocate bandwidth to those priority groups as a percentage of the maximum bandwidth available on the link.
Practically, this means that the port hardware must be able to support several queues for several class of traffic, identified by Typically, the FC and regular IP traffic is assigned to different priority groups, as shown in Figure 8. A bridge supporting ETS must support between two and eight groups, with at least one of them having no bandwidth limit. FC creates a lossless fabric using buffer-to-buffer credit see Figure 9. On a point-to-point FC link, a port can transmit data only if the remote peer has explicitly acknowledged with a buffer credit that it has the buffer capacity to receive this data.
Ethernet uses the opposite logic. The receiving port can issue a pause frame to stop the transmission from the remote peer when its buffer is about to be exhausted see Figure Note that the pause must be generated soon enough so that the buffer is not exhausted while the feedback mechanism is shutting down the remote port transmission. The pause mechanism, as defined more than ten years ago in This is not adapted to the goals of unified fabric, as mentioned in the previous section.
For example, if FC is exhausting its bandwidth cap and needs to be paused, the IP traffic has no reason to be impacted. An additional field within the pause frame specifies which priorities as defined with This mechanism allows creating both drop and no drop classes of traffic on the link. For example, because FC requires a lossless Ethernet, it is mapped to a priority group making use of the pause mechanism to stop data transmission before a drop can occur.
Figure 11 shows a pause frame that applies only to the priority group to which FC is mapped. IP traffic belonging to a different priority group is not affected. Note that IP does not require a lossless Ethernet; in case of congestion, the overflowing frames are simply dropped without generating a pause. Note However, there is already a consensus on the format of the pause format.
No significant change is expected at that stage. To minimize the configuration overhead, IEEE For example, devices may support a different number of priority groups, so discrepancies need to be detected. A server can be configured this way by the network when connected.
These functionalities are expected to be adopted in the IEEE standard. Introducing FCoE at the edge of the network does not require any change in the configuration of the FC director or the aggregation layer of the data center, so this document does not repeat the design recommendations above the access layer.
This form of redundancy is possible only because the FC nodes have a lot of intelligence built-in to their adapter and are tightly interacting with the services provided by the fabric. The fabric informs a particular node of the reachability of others. This allows the node to balance loads across fabrics or failover from one fabric to the other. Meeting the requirement for two strictly independent fabrics is not straightforward in an end-to-end converged approach, as shown in the right side of Figure 1.
However, when keeping FCoE at the edge, this function is still preserved. Figure 12 shows a logical view of the test network used for this document.
The Cisco Nexus s in Figure 12 are also used in the following ways:. This hybrid setup was not chosen as a design recommendation, but rather as a way of showing the following:.
This section details the configuration steps used to connect the access FC switch to the core directors. Whether the access switch is a Cisco Nexus or a conventional FC switch does not matter here. Cisco NX-OS Software provides exactly the same interface for configuration and maintenance in both cases, because native FC uplinks are used in the scenario described in this document.
These two options imply different configuration on the director and access switch, as detailed in the following sections. In switch mode, as shown in Figure 13 , the ports between the director and access switches are expansion ports E ports. Traffic between N ports is directly switched by the access FC switch.
Because of this function, this mode is especially recommended when there is a mix of initiators and targets at the access. It is furthermore best practice to configure the redundant connections between the access and the director as a channel.
Channeling provides granular load balancing between the uplinks as well as minimal disruption to the fabric if one link fails. The main drawback of the switch mode is that the access switch is assigned a domain ID.
FIP is used to perform device discovery, initialization, and link maintenance. FIP performs the following protocols:. The FCoE devices use these messages to perform a fabric login. That packet has a 4-bit version field. A Fibre Channel frame consists of 36 bytes of headers and up to bytes of data for a total maximum size of bytes. The encapsulated Fibre Channel frame has all the standard headers, which allow it to be passed to the storage network without further modification.
The IEEE Frames that are not correctly tagged are discarded. The switch will still accept untagged frames from the CNA. The packet has a 4-bit version field. The following table describes the FIP operation codes. DCBX end points exchange request and acknowledgment messages.
For flexibility, parameters are coded in a type-length-value TLV format. When an Ethernet interface is brought up, the switch automatically starts to communicate with the CNA. The CNAs that are connected to a Cisco Nexus Series switch are programmed to accept the configuration values sent by the switch, allowing the switch to distribute configuration values to all attached CNAs, which reduces the possibility of configuration errors and simplifies CNA administration.
The switch and CNA exchange capability information and configuration values. The Cisco Nexus Series switches support the following capabilities:. If a capability and its configuration values match between the switch and the CNA, the feature is enabled. If the CNA is configured to accept the switch configuration value, the capability is enabled using the switch value.
If the CNA is not configured to accept the switch configuration value, the capability remains disabled. On a per-interface basis, you can force capabilities to be enabled or disabled. Standard Ethernet is a best-effort medium which means that it lacks any form of flow control. In the event of congestion or collisions, Ethernet will drop packets. The higher level protocols detect the missing data and retransmit the dropped packets.
On a native Fibre Channel link, some configuration actions such as changing the VSAN require that you reset the interface status. When you reset the interface status, the switch disables the interface and then immediately reenables the interface. If an Ethernet link provides FCoE service, do not reset the physical link because this action is disruptive to all traffic on the link.
In this case, all traffic on the Ethernet interface is disrupted. A FIP adapter uses the FIP to exchange information about its available capabilities and to negotiate the configurable values with the switch. A pre-FIP adapter uses DCBX to exchange information about its available capabilities and to negotiate the configurable values with the switch. To reduce configuration errors and simplify administration, the switch distributes the configuration data to all the connected adapters.
The following figure shows a best practices topology for an access network using directly connected CNAs with Cisco Nexus Series switches. You must configure the unified fabric UF links as trunk ports. The following figure shows a best practices topology for an access network using remotely connected CNAs with Cisco Nexus Series switches. A blade switch must connect to exactly one Cisco Nexus Series converged access switch, preferably over an EtherChannel, to avoid disruption due to STP reconvergence on events such as provisioning new links or blade switches.
You must configure the Cisco Nexus Series converged access switch with a better STP priority than the blade switches that are connected to it.
This requirement allows you to create an island of FCoE VLANs where the converged access switch is the spanning-tree root and all the blade switches connected to it become downstream nodes.
0コメント