Ethernet Support Guide for HP-UX 11i v3

Table 3 Description of features
DescriptionFeature
Allows use of flow control negotiation and the sending and receiving of pause frames. For
information about configuring flow control, see Section 3.6 (page 27).
Receive and transmit
flow control
The CKO option enables the network adapter to compute the TCP checksum on transmit and
receive, saving the host from having to compute the checksum. This feature reduces CPU
Transmit and receive
Checksum Offload
(CKO) overhead. Some driver interfaces have an attribute that allows you to enable or disable receive
and send CKO; see Section 3.9 (page 32). The feature is valid with IPv4 only.
The TSO option breaks down large groups of data sent over the network into smaller segments.
(Also called large segment offload, LSO.) This feature reduces CPU overhead. Some driver
Transmit TCP
segmentation offload
(TSO) interfaces have an attribute that allows you to enable or disable receive and send CKO and
transmit TSO; see Section 3.10 (page 34). The feature is valid with IPv4 only.
The reverse of segmentation. Segmented units of data are reassembled into a stream of data
in its original form. (Also called large receive offload, LRO.) Some driver interfaces have an
TCP segment
reassembly in driver
attribute that allows you to turn TCP segment reassembly on or off. For more information, see
Section 3.11 (page 35).
HP Flex-10 allows each 10 Gigabit Ethernet adapter port to be divided into up to four
individual FlexNICs with bandwidth allocation in 100 Mb/s increments (to the maximum of
Flex-10 bandwidth
allocation
10 Gb/s per adapter port). Each FlexNIC is an abstraction of a portion of the Virtual Connect
connection, represented to the O/S as a standard NIC — each FlexNIC has a unique MAC
address and supports key benefits of integrated switching, including port aggregation, failover,
and VLAN tagging. For more information about HP Virtual Connect, see http://
h18004.www1.hp.com/products/blades/virtualconnect/index.html.
Network traffic streams are distributed into queues that can be associated with specific
processor cores. This distributes the workload and prevents data traffic processing from
Multiple receive queues
overwhelming a single core, for example. The packet queues can be accessed by driver
threads running on different processor cores, such that multiple cores can process network
packets in parallel. Some driver interfaces have an attribute that allows you to configure
multiple queues. For more information, see Section 3.8 (page 31).
See description of multiple receive queues. For more information, see Section 3.8 (page 31).Multiple transmit
queues
Also known as packet steering on the receive side, when multiple receive queues are supported
by the driver and NIC, the NIC steers inbound packets to different queues so that the network
Receive Side Scaling
(RSS) with TCP
load is shared across multiple CPUs to avoid a single-processor bottleneck. RSS enables
packet receive-processing to scale with the number of available processors.
UDP 4-tuple uniquely describes a User Datagram Protocol socket pair. TCP 4-tuple uniquely
describes a Transmission Control Protocol socket pair. The 4-tuple is a set of source and
UDP 4-tuple, TCP
4-tuple
destination IP addresses and port numbers (local and remote socket addresses) that are used
for calculations. In contrast, a 2-tuple only has source and destination IP addresses.
Large 9000–byte maximum transmission unit (MTU) for improved efficiency and performance
for bulk data transfers. Some driver interfaces have an attribute that allows you to configure
Jumbo Frames; see Section 3.7 (page 29).
Jumbo Frames
In multicast mode one or more multicast group addresses are programmed into the network
card through the driver; the card receives and processes all multicast packets corresponding
Multicast and
Promiscuous mode
to the programmed addresses. In contract, promiscuous mode allows all packets to be received
and processed, regardless of type or addressee.
Internet protocol version 6, a version of IP currently used to direct almost all Internet traffic.
Intended to succeed IPv4.
IPv6
The IEEE standard supports VLANs on an Ethernet network, defining a system of VLAN tagging
and stripping of tags for Ethernet frames, and the accompanying procedures used by bridges
IEEE802.1Q VLAN
tagging and stripping
in h/w and switches in handling such frames. For more information about HP-UX VLAN functionality,
see http://www.hp.com/go/vlan.
Provides a flexible mechanism for managing CPU interrupt assignments by moving external
I/O device interrupts from one CPU to another to prevent performance degradation that would
occur if, for example, two heavily loaded I/O cards mapped interrupts to the same CPU.
Interrupt migration
10 Overview