How a cluster works
Cluster management is implemented through HW Group Management Protocol version 2 (HGMPv2), which consists of the following three protocols:
Neighbor Discovery Protocol (NDP)
Neighbor Topology Discovery Protocol (NTDP)
Cluster
A cluster configures and manages the devices in it through the above three protocols. Cluster management involves topology information collection and the establishment and maintenance of a cluster. Topology information collection and cluster maintenance are independent from each other, with the former starting before the cluster is created:
All devices use NDP to collect the information of the directly connected neighbors, including their software version, host name, MAC address and port number.
The management device uses NTDP to collect the information of the devices within user-specified hops and the topology information of all devices, and then determines the candidate devices of the cluster based on the collected information.
The management device adds or deletes a member device and modifies cluster management configuration according to the candidate device information collected through NTDP.
Introduction to NDP
NDP is used to discover the information about directly connected neighbors, including the device name, software version, and connecting port of the adjacent devices. NDP works in the following ways:
A device running NDP periodically sends NDP packets to its neighbors. An NDP packet carries NDP information (including the device name, software version, and connecting port, etc.) and the holdtime, which indicates how long the receiving devices will keep the NDP information. At the same time, the device also receives (but does not forward) the NDP packets from its neighbors.
A device running NDP stores and maintains an NDP table. The device creates an entry in the NDP table for each neighbor. If a new neighbor is found, meaning the device receives an NDP packet sent by the neighbor for the first time, the device adds an entry in the NDP table. If the NDP information carried in the NDP packet is different from the stored information, the corresponding entry and holdtime in the NDP table are updated; otherwise, only the holdtime of the entry is updated. If no NDP information from the neighbor is received when the holdtime times out, the corresponding entry is removed from the NDP table.
NDP runs on the data link layer, and therefore supports different network layer protocols.
Introduction to NTDP
NTDP provides information required for cluster management; it collects topology information about the devices within the specified hop count. Based on the neighbor information stored in the neighbor table maintained by NDP, NTDP on the management device advertises NTDP topology-collection requests to collect the NDP information of all the devices in a specific network range as well as the connection information of all its neighbors. The information collected will be used by the management device or the network management software to implement required functions.
When a member device detects a change on its neighbors through its NDP table, it informs the management device through handshake packets. Then the management device triggers its NTDP to collect specific topology information, so that its NTDP can discover topology changes timely.
The management device collects topology information periodically. You can also administratively launch a topology information collection. The process of topology information collection is as follows:
The management device periodically sends NTDP topology-collection request from the NTDP-enabled ports.
Upon receiving the request, the device sends NTDP topology-collection response to the management device, copies this response packet on the NTDP-enabled port and sends it to the adjacent device. Topology-collection response includes the basic information of the NDP-enabled device and NDP information of all adjacent devices.
The adjacent device performs the same operation until the NTDP topology-collection request is sent to all the devices within specified hops.
When the NTDP topology-collection request is advertised in the network, large numbers of network devices receive the NTDP topology-collection request and send NTDP topology-collection response at the same time, which may cause congestion and the management device busyness. To avoid such case, the following methods can be used to control the speed of the NTDP topology-collection request advertisement:
Upon receiving an NTDP topology-collection request, each device does not forward it, instead, it waits for a period of time and then forwards the NTDP topology-collection request on the first NTDP-enabled port.
On the same device, except the first port, each NTDP-enabled port waits for a period of time and then forwards the NTDP topology-collection request after its prior port forwards the NTDP topology-collection request.
Cluster management maintenance
Adding a candidate device to a cluster
You should specify the management device before creating a cluster. The management device discovers and defines a candidate device through NDP and NTDP protocols. The candidate device can be automatically or manually added to the cluster.
After the candidate device is added to the cluster, it can obtain the member number assigned by the management device and the private IP address used for cluster management.
Communication within a cluster
In a cluster the management device communicates with its member devices by sending handshake packets to maintain connection between them. The management/member device state change is shown in Figure 85.
Figure 85: Management/member device state change
After a cluster is created, a candidate device is added to the cluster and becomes a member device, the management device saves the state information of its member device and identifies it as Active. And the member device also saves its state information and identifies itself as Active.
After a cluster is created, its management device and member devices begin to send handshake packets. Upon receiving the handshake packets from the other side, the management device or a member device simply remains its state as Active, without sending a response.
If the management device does not receive the handshake packets from a member device in an interval three times of the interval to send handshake packets, it changes the status of the member device from Active to Connect. Likewise, if a member device fails to receive the handshake packets from the management device in an interval three times of the interval to send handshake packets, the status of itself will also be changed from Active to Connect.
If this management device, in information holdtime, receives the handshake or management packets from its member device which is in Connect state, it changes the state of its member device to Active; otherwise, it changes the state of its member device to Disconnect, in which case the management device considers its member device disconnected. If this member device, which is in Connect state, receives handshake or management packets from the management device in information holdtime, it changes its state to Active; otherwise, it changes its state to Disconnect.
If the communication between the management device and a member device is recovered, the member device which is in Disconnect state will be added to the cluster. After that, the state of the member device locally and on the management device will be changed to Active.
Besides, a member device informs the management device using handshake packets when there is a neighbor topology change.
Management VLAN
The management VLAN is a VLAN used for communication in a cluster; it limits the cluster management range. Through configuration of the management VLAN, the following functions can be implemented:
Management packets (including NDP, NTDP and handshake packets) are restricted within the management VLAN, therefore isolated from other packets, which enhances security.
The management device and the member devices communicate with each other through the management VLAN.
For a cluster to work normally, you must set the packets from the management VLAN to pass the ports connecting the management device and the member/candidate devices (including the cascade ports). Therefore:
If the packets from the management VLAN cannot pass a port, the device connected with the port cannot be added to the cluster. Therefore, if the ports (including the cascade ports) connecting the management device and the member/candidate devices prohibit the packets from the management VLAN, you can set the packets from the management VLAN to pass the ports on candidate devices with the management VLAN auto-negotiation function.
Only when the default VLAN ID of the cascade ports and the ports connecting the management device and the member/candidate devices is that of the management VLAN can you set the packets without tags from the management VLAN to pass the ports; otherwise, only the packets with tags from the management VLAN can pass the ports.
NOTE:
If a candidate device is connected to a management device through another candidate device, the ports between the two candidate devices are cascade ports.
For more information about VLAN, see Layer 2—LAN Switching Configuration Guide.