IBM PureFlex System and IBM Flex System Products and Technology

263

Enterprise Chassis, create a loop in a Layer 2 network. For more information, see Topology 2

in Figure 6-4 on page 262. In this case, use STP as a loop prevention mechanism because a

Layer 2 network cannot operate in a loop.

Assume that the link between TOR 2 and Enterprise Chassis switch 1 is disabled by STP to

break a loop. Therefore, traffic goes through the link between enterprise switch 1 and

Enterprise Chassis switch 1. During link failure, STP reconfigures the network and activates

the previously disabled link. The process of reconfiguration can take tenths of a second,

during which time the service is unavailable.

Whenever possible, plan to use trunking with VLAN tagging for interswitch connections. This

configuration can help achieve higher performance by increasing interswitch bandwidth, and

higher availability by providing redundancy for links in the aggregation bundle. For more

information about trunking, see 6.5.1, “Trunking” on page 266.

STP modifications, such as Port Fast Forwarding or Uplink Fast, might help to improve STP

convergence time and the performance of the network infrastructure. Additionally, several

instances of STP can run on the same switch simultaneously. These instances run on a

per-VLAN basis. That is, each VLAN has its own copy of STP to load balance traffic across

uplinks more efficiently.

For example, assume that a switch has two uplinks in a redundant loop topology and several

VLANs are implemented. If single STP is used, one of these uplinks is disabled and the other

carries traffic from all VLANs. However, if two STP instances are running, one link is disabled

for one set of VLANs while carrying traffic from another set of VLANs, and vice versa. In other

words, both links are active, thus enabling more efficient use of available bandwidth.

6.4.3 Layer 2 failover

Each compute node can have one IP address per each Ethernet port, or one virtual NIC

consisting of two or more physical interfaces with one IP address. This configuration is known

as NIC teaming technology. From the Enterprise Chassis perspective, NIC Teaming is useful

when you plan to implement high availability configurations with automatic failover in case of

internal or external uplink failures.

You can use only two ports on compute node per virtual NIC for high availability

configurations. One port is active, and the other is standby. One port is connected to a switch

in I/O bay 1, and the other port to a switch in I/O bay 2. If you plan to use an Ethernet I/O

Adapter for high availability configurations, the same rules apply. Connect the active and

standby ports to switches on different bays.

During internal port or link failure of the active NIC, the teaming driver switches the port roles.

The standby port becomes active and the active port becomes standby. This process takes

only a few seconds. After restoration of the failed link, the teaming driver can run a failback or

do nothing, depending on the configuration.

Look at topology 1 in Figure 6-4 on page 262. Assume that NIC Teaming is on, and that the

compute node NIC port connected to switch 1 is active and the other is on standby. If

something goes wrong with the internal link to switch 1, the teaming driver detects the NIC

port failure and runs a failover. If external connections are lost, such as the connection from

Enterprise Chassis switch 1 to top of rack switch 1, nothing happens. There is no failover

because the internal link is still on and the teaming driver does not detect any failure.

Therefore the network service becomes unavailable.

To address this issue, use the Layer 2 Failover technique. Layer 2 Failover can disable all

internal ports on switch module in the case of an upstream links failure. A disabled port