feature. Figure 3 illustrates a dual host system with an active
and standby processor configuration. The redundancy of the
host and the fabric can be achieved through the NTB function
of PEX 8609. Additionally, the DMA engine in the PEX8609
can be used to mirror the data on both the active and standby
hosts.
Figure 5. Distributed Computing
Intelligent Adapter Card Redundancy
The PEX 8609 supports the
non-transparency
feature.
Figure 6 illustrates a host system using an intelligent adapter
card.
Figure 6. Intelligent Adapter Usage
Figure 3. Active-Standby Failover
In this figure, the CPU on the adapter card is isolated from the
host CPU. The PEX 8609 non-transparent port allows the two
CPUs to be isolated but communicate with each other through
various registers that are designed in the PEX 8609 for that
purpose. Moreover, the internal DMA engine can be used to
copy or mirror the data between adapter cards so that
intelligent I/O adapter redundancy is achieved.
This model can be expanded to an Active-Standby failover
model. In this case, there is an Active host and a Standby host
ready to take over the system in the event the Active host fails.
The DMA engine in the PEX8609 can be used to mirror the
data between hosts thus minimizing the failover time.
DMA in Control Planes
The DMA engine in the PEX8609 can alternatively be used to
perform housekeeping tasks, such as reading device status
information from the many endpoints available. This
particular usage model allows the DMA engine to perform the
many data transfers for endpoint status that would otherwise
be handled by the processor.
Generic Data Mover
The DMA engine in the PEX8609 is very flexible. It
implements
Four DMA channels
capable of saturating a x8
Gen2 PCI Express link. Furthermore, it supports a Descriptor
Ring approach in which the descriptors can be placed in
external host memory or internal to the PEX8609. The DMA
engine can move data from any port to any port including the
same port. That is, the source and destination for a DMA
transfer can be on the same PEX8609 port. Figure 7 shows
two PCIe endpoints and a PEX8609 device connected through
a PCIe switch. In this example, one PCIe endpoint behind the
PEX8609 takes advantage of the DMA capabilities in the
PEX8609 to move data from PCIe2 to PCIe1.
Figure 4. PEX8609 DMA in a Packet Router
Distributed Computing
The PEX 8609 is well suited for distributed computing
applications. Once programmed, the DMA engine in the
PEX8609 transfers the data between servers and through the
NT port at a high rate. The processor can focus all of its
computing cycles manipulating the large amounts of data
received.
Figure 7. Generic Data Mover
PLX [ PLX TECHNOLOGY ]
