Shulou(Shulou.com)06/03 Report--

In order to improve throughput and redundancy, netapp storage aggregates multiple network connections to provide services to a single logical port group. There are 3 aggregation modes in the port group of netapp, which are:

Single mode

Static Multimode

Dynamic Multimode

The working principle of the mode in 3 and the configuration example of the switch side are described in detail below.

Single mode

In a single mode port group, no matter how many ports there are, only one port is active, the rest are dormant / standby, only the active port fails, and the storage system randomly activates one to take over communication from the hibernation / standby port.

In this port group, all network interfaces share a single MAC address.

The state detection and failover of the entire port group are all done by the storage host, and the switch is not required to participate, so the switch does not need to be configured. However, switch ports linked to links in the same port group must be in the same broadcast domain (in the same subnet or VLAN) and cannot span multiple switches (unless the switches are stacked).

The main features of Single mode:

The configuration is simple, and the switch side does not need to be configured.

Ability to provide network redundancy

Unable to improve network throughput.

This method is recommended when the network throughput is not a bottleneck.

As shown in the figure above, the single mode port group composed of two links is normally activated by e0 and standby by E1. If e0 fails, then E1 is activated.

Static Multimode

Static multimode is actually compatible with link aggregation but does not support LACP (Link Aggregation Control protocol) network mode. All switches that support EEE 802.1AX-2008 (formerly known as IEEE 802.3ad) can be used with data ontap's Static Multimode port group mode.

Note: LACP is not a link aggregation link aggregation, but a subprotocol in link aggregation. Supporting LACP is dynamic link aggregation, otherwise it is static link aggregation (static multimode). Some devices only support dynamic link aggregation, such as cisco UCS fabric interconnect only supports dynamic link aggregation

The following conditions need to be met to implement this pattern:

All ports are in full duplex mode

The port rate is the same in the port group

All uplink physical ports need to be on one switch or on the same stacked switch

The features are as follows:

Ethernet frames are allowed to be distributed across physical links, so the throughput may exceed a single physical link. However, IEEE does not specifically define the distribution algorithm. But the guiding principle is: 1) for each specific session, the order of data frames cannot be disrupted; 2) there can be no duplicate data frames. Based on the above principles, that is, for a specific session, all data frame transmission is carried out on a physical link. That is, the upper limit of the transmission speed of a single session is the speed of a physical link. Although IEEE does not define specific algorithms, each vendor has developed a variety of algorithms.

All ports in the same aggregation group are active and share the same MAC address.

A port group consisting of n physical links can tolerate nmur1 link failures. When one link fails, the traffic transmitted on the failed link will automatically not be assigned to other healthy links, and when the failed link is restored, the traffic will be automatically redistributed on all active links.

Dynamic Multimode

Similar to static multimode, but supports LACP. All ports of the same port group share a single MAC address

Note: data Ontap does not support Cisco's private protocol Port Aggregation Protocol (PAgP). In addition, Data Ontap uses LACP's active mode, so the switch side can configure LACP active/passive mode at will

Configuring Dynamic multimode requires the following to be configured at the same time:

LACP also does not define a load balancing algorithm. Therefore, the load balancing strategy is also needed on the switch that stores the uplink. Such as IP-based, MAC-based, or round robin.

Static vs Dynamic multimode

Physical link failures can be detected in both modes.

Because there is no LACP control protocol to negotiate, the former cannot detect logic errors in the configuration. For example, one end is configured correctly, and the misconfiguration of one side may lead to unpredictable results, such as MAC address storms.

The latter can detect some errors in the type of non-physical link loss, such as inconsistent parameter configuration at both ends. This mode provides more reliable link aggregation.

If all devices support it, Dynamic multimode is recommended.

Load balancing algorithm Round robin

Based on this strategy, data frames will be evenly distributed to each active link for transmission, such as data frame 1 through link1, data frame 2 through link2, and so on. However, there may be a risk that the data frames in a session may not arrive in order. For example, data frame 1 is sent over link1 and data frame 2 is sent over link2. Normally, data frame 1 should arrive before data frame 2, but due to link1 congestion, it may cause data frame 1 to arrive after data frame 2. In this way, the application side may need to retransmit the data frame.

Based on source and destination pairing strategy

This is the most frequently used policy type for Netapp Multimode. It also includes: source and destination MAC address, source and destination IP address. This type of load balancing method performs XOR operations on the information of the source and destination, then divides it by the number of physical links in the aggregation, and finally calculates the physical links selected for data frame transmission. The core of the algorithm is that each calculated result corresponds to a physical link. So, here's a concept to clarify. Many people think that the maximum bandwidth that can be obtained after four 1Gb links are aggregated is 4Gb. This is actually wrong, and the maximum bandwidth of a single transmission is actually 1Gb, because according to the source and destination algorithms, the transmission always uses only one physical link.

Source and destination MAC address

Select the physical link based on the MAC address of the source side (such as a PC) and the destination side (storage controller). If the source side and the destination side are on the same subnet, the final non-uniformity of all traffic will not be very prominent. However, if the source and destination are not on the same subnet, they need to cross the router. Then there may be a serious traffic imbalance. For example, if there are 100hosts in subnet1 and now need the storage controller of subnet2, then the source MAC address of the last 100hosts will be the MAC of the router. In this case, the communication between the hosts and the controller will all choose the same physical link. It is very likely that this link will cause congestion, while other links will be very idle.

Source and destination IP address

What you need to understand is that under this computing strategy, getting is not all of source and destination IP, but only the last bit of a decimal IP address. Such as host1 (10.0.0.100), host2 (10.1.1.100), controller C1 (10.2.2.1).

So when you run XOR, host1 XOR C1s 100 XOR 1 switch host2 XOR C 1s 100 XOR 1

The calculated values are the same.

Example of Cisco Catalyst switch configuration

Cisco's link aggregate technology is etherchannel.

Configure layer 2 etherchannel

Step1: enter global configuration mode

Configure terminal

Step2: specifies the physical port to join the aggregation group

Interface interface-id

Step3: assign VLAN to the port or configure it as trunk

Switchport mode {access | trunk}

Switchport access vlan vlan-id

Step4: enable port fast convergence

Spanning-tree portfast

Step5: adds the port to the specified aggregation group and specifies the aggregation mode

Channel-group channel-group-number mode {auto [non-silent] | desirable [non-silent] | on} | {active | passive}

On: if the Netapp side is configured as Static multimode, then the cisco switch adopts this mode

Active | passive: if the netapp side is configured as the Dynamic multimode,cisco switch, the mode is configured.

Step 6: configure load balancing policy

Port-channel load-balance src-dst-ip

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.