Configuration Guide Vol. 1


7.7.2 Notes on stacks

<Structure of this section>

(1) Operation of the configuration file

(2) Configuration that requires a device or VLAN programming restart

If you have edited a configuration that requires a reboot of the appliance or VLAN programme to reflect any changes you have made, you must reboot all member switches. Edit the configuration, save the changes to the startup configuration using the save command, and then restart all the member switches. For the reboot procedure, see 8.2.6 Restarting the Stack.

The applicable configuration commands are listed below.

Of these, ip route static maximum-paths,ipv6 route static maximum-paths, and maximum-paths commands only need to be restarted when an alert-level operation message is issued after editing the configuration. For more information, see "Configuration Guide Vol.3" "8.4.2 Load Balancing Specifications".

If you reboot and operate only the member switch whose configuration has been changed without restarting all the member switches, the new configuration will be applied only to the restarted member switches.

For example, if you use the following configuration command to change the configuration of a table entry, and only the member switch whose configuration was changed is restarted, the table entry operates in a different state for each member switch.

At this time, the number of table entries that are guaranteed to work is limited to the minimum of all member switches. To check the table entries for a member switch, execute the operation-command show system.

(3) About forwarding packets when IPv4 multicast is used

When IPv4 multicast is used on a stack, when a multicast forwarding entry is changed, packets that are forwarded to the corresponding forwarding entry may be discarded without forwarding Layer 2. When the negative cache for multicast forwarding is changed, Layer 3 discarded packets in the corresponding negative cache may be discarded without forwarding them at Layer 2.

(4) About Flow Control

Flow control does not work on stacking ports.

If a stack uses flow control and a member switch runs out of receive buffers, other member switches might not run out of buffers. As a result, if a member switch is stuck in a transmit packet and the receive buffer is exhausted, no pause packet is sent from another member switch.

(5) About MAC address-learning

In the stack, every member switch learns its own MAC addressing. After a member switch learns MAC address, it takes up to 180 seconds for the learned MAC address to be reflected in the other member switches. For stable operation of MAC address learning, Xilinx recommends that you do not reduce the aging time of learned MAC addresses below the default of 300 seconds.

Because individual member switches learn MAC addressing, there are two limitations:

(a) Limitations on detecting MAC address-learning moves

When a terminal, such as a PC, is moved from one member switch port to another, the member switch to which the terminal moved detects the movement and reflects the learned MAC address in MAC address table of each member switch. However, depending on the number of moves and the frequency of the moves, the following problems occur:

  • If many terminals move at the same time, MAC address table of the member switches may contain MAC addresses learned on the port before the move, except for the move destination. In this state, because the frame is sent to the port before moving, normal communication may not be possible.

  • If MAC addresses are learned up to the capacity limit of MAC address table, if many terminals move frequently, MAC addresses learned by the member switches may not be reflected in the other member switches within the above-mentioned period. When this happens, frames that are destined for MAC addresses that have not been reflected are flooded.

In such cases, wait until the newly learned MAC on the member switches is reflected on the other member switches.

(b) Restricting Unicast Communication

If two terminals are connected to different member switches, unicast communication from either terminal may be flooded into VLAN even if unicast communication is performed between the two terminals. In this case, wait until either of the following conditions is satisfied:

  • Multicast or broadcast packets are sent from the terminal that is the destination of the flooded frame

  • MAC addresses learned by the member switches are reflected in other member switches.

(6) Diverting Member Switches Used in the Stack

In a stack, the device MAC address is the chassis MAC address of the master switch when the stack is configured for the first time. After that, even if the master switch fails, the device MAC is not changed.

Therefore, when removing a member switch used in the stack from the stack and connecting this equipment to the same network as the corresponding stack, make sure that the device MAC address of the stack differs from the chassis MAC address of the removed member switch in advance. In the same case, change the device MAC of the stack by removing the corresponding member switch from the stack and then restarting the stack.

For details about the device MAC address of the stack, see 7.3.5 Device MAC address of the stack and 8.2.6 Restarting the stack.

(7) Switching the master switch

To switch the master switch while the packet transfer continues, confirm that both of the following conditions are satisfied before switching.

If you switch the master switch while the backup switch is initializing, the backup switch during initialization restarts and cannot continue forwarding packets.

You can use the operation-command show switch to check whether the backup switch has been initialized. You can also use the operation-command show port to check whether the port is up.

(8) When using master election priority 1

When a stack consisting of one member switch with master election priority 1 is connected to a stack consisting of one member switch with master election priority 2 or higher, a member switch with master election priority 2 or higher is selected as the master switch. The master election priority 1 member switch restarts and joins the stack as a backup switch. A member switch with master election priority 2 or higher continues in the master state without restarting, so the stack transfer function is maintained.

However, when a member switch with master election priority 2 or higher is connected to a stack consisting of one member switch with master election priority 1 and started up, the member switch with master election priority 2 or higher waits for the backup transition instruction from the master switch in the initial state to detect the master switch. At the same time, the member switch with master election priority 1 restarts to detect the member switch with master election priority 2 or higher. A member switch waiting for a backup transition instruction detects the absence of the master switch and restarts it. During this time, communication is interrupted until a member switch with master election priority 2 or higher completes initialization as a master switch.

In this way, if master election priority 1 is used, communication interruption time may become longer when the master switch is switched.

When adding a member switch to an existing stack, master election priority 1 should be used for temporary operations to prevent unintended configuration replacements. In normal operation, it is not recommended to use master election priority 1 to fix the master switch election. After stack construction, it is recommended to set master election priority 2 or higher.

(9) About member switch startup time when storm control is used

If the stack uses storm control with a threshold for the number of received frames, adding a member switch will take several minutes longer than without storm control. The addition of a member switch means the following:

(10) Handling of management ports in stack configuration

The management port operates only on the master switch, and the management port of the backup switch is shut down.

The management port is enabled on the switch that is switched from the backup switch to the master switch. Because MAC address of the management port is changed, communication cannot be performed until the address information of the operation terminal is updated.

(11) Combinations of Ports That Can Attach Stack Links

The stack link has a limit on the combination of ports and connection interfaces that can be connected. Operation cannot be guaranteed if the device is used in a combination that is not supported.

Table 7-7: Combinations of ports that can connect stack links (connections in 40GBASE-R)

Connection interface

Master switch

Backup switch

Port type

Port number

QSFP + port #1

QSFP28/QSFP+

Shared port #2

Port 49~50

QSFP28/QSFP+

Shared port #2

Port 51~52

40GBASE-SR4

40GBASE-LR4

QSFP + port #1

OK

OK

OK

QSFP28/QSFP+

Shared port #2

Port 49~50

OK

OK

OK

QSFP28/QSFP+

Shared port #2

Port 51~52

OK

OK

OK

40GBASE-CR4

QSFP + port #1

OK

OK

NG

QSFP28/QSFP+

Shared port #2

Port 49~50

OK

OK

NG

QSFP28/QSFP+

Shared port #2

Port 51~52

NG

NG

OK

Legend OK: Supported; NG: Not supported

Note #1

AX3660S-24T4X,AX3660S-24T4XW,AX3660S-48T4XW,AX3660S-16S4XW, and AX3660S-24S8XW

Note #2

AX3660S-48XT4QW,AX3660S-24X4QW, and AX3660S-48X4QW

Table 7-8: Combinations of ports that can connect stack links (connections in 100GBASE-R)

Connection interface

Master switch

Backup switch

Port type

Port number

QSFP28/QSFP+

Shared port*

Port 49~50

QSFP28/QSFP+

Shared port*

Port 51~52

100GBASE-SR4

100GBASE-CWDM4

100GBASE-LR4

100GBASE-4WDM-40

QSFP28/QSFP + shared port*

Port 49~50

NG

NG

QSFP28/QSFP + shared port*

Port 51~52

NG

NG

100GBASE-CR4

QSFP28/QSFP + shared port*

Port 49~50

OK

NG

QSFP28/QSFP + shared port*

Port 51~52

NG

OK

Legend OK: Supported; NG: Not supported

#: AX3660S-48XT4QW,AX3660S-24X4QW, and AX3660S-48X4QW