Failover (Network cards)

Failover is an MDI feature that allows networking traffic to be shifted to a different configured network adapter card when hardware failures are detected on the original card.

Three types of failover support are available for SVR5 MDI devices:

PCI hotplug failover

Requires hardware that implement PCI hotplug failover (for example, Compaq Proliant 6500, Proliant 7000, and IBM) and a DDI 8 MDI driver that implements the CFG_SUSPEND and CFG_RESUME subfunctions to the config(D2mdi) entry point routine. This allows the system administrator to suspend the driver, power down the slot, hot-swap or hot-replace the network adapter card, power up the slot, and resume the driver, all without bringing the machine down. Your driver does not need to send up MAC_HWFAIL_IND(D7mdi) messages. For MDI versions 2 and 2.1, this works only with DDI 8 and later drivers for PCI adapters.

Reliant non-stop clustering

Requires that your driver be able to program MAC addresses and its own address with the MACIOC_SETMCA and MACIOC_SETADDR ioctls. This works with any DDI 7 or DDI 8 MDI driver on any bus type.

General failover

The DLPI subsystem listens for MAC_HWFAIL_IND(D7mdi) messages and initiates a failover if the failover criteria are met. The system administrator has the ability to set failover criteria: for some sites, a single MAC_HWFAIL_IND message may be an indication to move all traffic to another failover network adapter card. Note that all the network adapter cards must already be present in the machine and that failover is an automatic process after it is initially configured; no user intervention is required to switch to a new card.

The Network Configuration Manager is used to configure a failover adapter. It also provides the ability to force a failover and to failback to the original adapter after the problem is resolved. This is useful for testing as well as administration.

With appropriate hardware, you can combine the hot-swap and general failover features. This enables the user to hot-swap the primary network adapter card after a failover, then failback to the replaced card.

General failover works with any MDI versions 2 or 2.1 driver on any bus type.

All MDI drivers should be written to support failover from the card. Criteria are:

• Driver must send up a MAC_HWFAIL_IND(D7mdi) message to indicate that it has problems. The driver for any device that can detect a fatal error should send up this message, whether or not it claims to support failover. This is not required for PCI hotplug devices.

• Driver must fully re-initialize the adapter across a close( ) followed by a subsequent open of the device. Remember that an MDI driver's close(D2mdi) routine must disable interrupts on the card and the driver's open(D2mdi) routine must fully initialize the card. If the card cannot be reset, the open( ) routine should fail. For MDI drivers, the /dev/mdi device is exclusive open.

Criteria for failover to your card:

• Driver must implement the MACIOC_SETADDR ioctl to set the MAC address.

• Driver must implement the MACIOC_SETMCA ioctl to set the multicast address.

• Driver must implement the MACIOC_SETALLMCA ioctl to receive all multicast addresses.

• Driver must implement the MACIOC_DELALLMCA ioctl to stop receiving packets sent to any multicast address.

• Driver must implement an ``infinite'' number of multicast addresses. In practice, this means that your hardware may use perfect address matching for some finite number of multicast addresses (such as 16, 32, 64, 246, depending on the hardware's capability). For any subsequent MACIOC_SETMCA commands the driver should enable imperfect address filtering if the hardware supports it, or enable the receipt of all multicast addresses. If either imperfect address filtering or all multicast reception is enabled, the driver should call the mdi_valid_mca(D3mdi) function for each subsequent multicast frame that is received. See ``Promiscuous mode''.

  Drivers should not go into promiscuous mode if the hardware is not able to enable the receipt of all multicast addresses or imperfect address matching (hashing).

• Drivers that implement all these requirements should set FAILOVER=true in their bcfg(DSP/4dsp) files so that netcfg can take appropriate actions to implement failover either from or to your card as required.

You can use the Network Configuration Manager to configure a failover device for your adapter and test failover as follows:

1. Generate traffic to another machine with the ping -s <remote> command.

2. Run nd failover net<X>, where net<X> is the net[0-9] device configured over your driver. ndgetmdi shows all configured net<X> devices associated with MDI drivers.