I/O domain resiliency improves the availability and performance of an I/O domain by enabling it to continue to run even when one of its associated root domains is interrupted. When a root domain is interrupted, the I/O domains that use its services continue to run by enabling its affected devices to fail over to the alternate I/O path. When the root domain returns to service, the affected devices in the resilient I/O domain are also returned to service and the failover capabilities are restored.
I/O domain resiliency is a logical domain configuration strategy that involves:
Creating NPRDs to provide virtualized I/O services to the I/O domain. The NPRDs own at least one PCIe bus from at least one CMIOU.
Creating virtualized I/O services from the NPRDs to the I/O domain. The I/O domain must have at least two I/O connections with one of the connections coming from one IOH on one CMIOU and the other connection coming from another IOH on a different CMIOU.
Creating alternate paths to the I/O connections by using multipathing.
I/O domain resiliency has these restrictions:
Multipath I/O configurations are required for uninterrupted I/O services.
Current support is limited to SR-IOV virtual functions, virtual network devices, and virtual storage devices that support I/O domain resiliency..
For this feature, follow these guidelines:
For fibre channel cards:
Generally, add cards in pairs for redundancy.
Spread the cards across IOHs.
Enable MPxIO in the service domain.
For NIC cards:
Generally, add cards in pairs for redundancy.
Use IEEE 802.3ad link aggregations in the service domain.
For InfiniBand cards:
Generally, add cards in pairs for redundancy.
Avoid sharing the PCI bus with other cards, if possible.