Q-Logic 8100 SERIES Manuel D’Utilisation
6–Configuring NIC Functionality in the Converged Network Adapter
Configuring the NIC in a Windows Environment
Configuring the NIC in a Windows Environment
6-10
FE0254601-00 A
checksum offload unless you are debugging a checksum computation
problem. TCP checksum offloading significantly reduced reduces host CPU
use when using jumbo frames.
problem. TCP checksum offloading significantly reduced reduces host CPU
use when using jumbo frames.
Stateless offload—QLogic 8100 Series Adapters support large send
offloading (LSO).LSO enables the Microsoft Windows TCP stack to send
one large block of data to the QLogic adapter, which then segments this
large block into multiple TCP packets. The LargeSendOffloadSupport
parameter is enabled by default and can be disabled on the Windows
Advanced property pages for the QLogic 10Gb PCI Ethernet adapter using
Windows Device Manager. For information about the
LargeSendOffloadSupport parameter, refer to
offloading (LSO).LSO enables the Microsoft Windows TCP stack to send
one large block of data to the QLogic adapter, which then segments this
large block into multiple TCP packets. The LargeSendOffloadSupport
parameter is enabled by default and can be disabled on the Windows
Advanced property pages for the QLogic 10Gb PCI Ethernet adapter using
Windows Device Manager. For information about the
LargeSendOffloadSupport parameter, refer to
.
Receive Side Scaling
QLogic 8100 Series Adapters supports receive side scaling (RSS) in Microsoft
Windows environments for nonoffloaded IP/TCP traffic. When a packet arrives on
a network interface, an interrupt is sent to the network driver. The network driver
then executes a deferred procedure call (DPC), which runs on the same CPU as
the interrupt. With RSS disabled, only one DPC can execute at a time. With RSS
enabled, up to four parallel DPCs can run on four different processors or cores,
which enables the simultaneous receive processing of incoming packets
(
Windows environments for nonoffloaded IP/TCP traffic. When a packet arrives on
a network interface, an interrupt is sent to the network driver. The network driver
then executes a deferred procedure call (DPC), which runs on the same CPU as
the interrupt. With RSS disabled, only one DPC can execute at a time. With RSS
enabled, up to four parallel DPCs can run on four different processors or cores,
which enables the simultaneous receive processing of incoming packets
(
).
Figure 6-4. Receive Side Scaling Concepts
With RSS enabled, the NIC driver implements a hash function to distribute
inbound packets across CPUs. In-order delivery is maintained by identifying the
flow. RSS provides the following advantages:
inbound packets across CPUs. In-order delivery is maintained by identifying the
flow. RSS provides the following advantages:
Parallel processing of inbound packets while maintaining in-order delivery
Load balancing of network processing across CPUs in an symmetrical
multiprocessing (SMP) system
multiprocessing (SMP) system
Cache locality