 |
» |
|
|
|
HP APA offers you a comprehensive solution to create
fast, highly available network server connections with minimal IT
support costs. HP APA enables this with four key benefits: Automatic link failure detection and recovery in case
of network failures. A link aggregate continues to operate as long
as there is at least one port operating. Scalable high-performance link aggregates using Fast
or Gigabit Ethernet and the HP APA load-balancing algorithms. See “Load Balancing” for more information. Fault management and isolation with the HP MIB Monitor
and nettl logging facilities. Lower IT costs with automated configuration and management
tools using the IEEE 802.3ad or PAgP standards and the intuitive HP
System Management Homepage (SMH) GUI.
This section describes the following features
of HP APA: Interoperability with HP Serviceguard
Link Aggregate | |
HP APA enables you to
combine 2 to 8 physical link ports (up to 32 for LACP mode) into one
link aggregate. This gives the link aggregation a theoretical bandwidth
of 8 times that of a single physical link (32 times for LACP mode).
A link aggregate has the following characteristics:
The physical ports in the
link aggregation use the same MAC address. The unique MAC address for a specific link aggregate is determined
by using the MAC address of one of the ports in the link aggregate.
All ports will use the same MAC address. When a physical port is removed from a link aggregate,
the port's MAC address is reset to its own MAC address. HP APA link aggregates can migrate the network traffic
from a failed physical link in the aggregate to the remaining operational
links in the aggregate. HP APA distributes the outbound network traffic across
the physical links in the link aggregation using a load balancing
algorithm. Effective APA load balancing requires many
simultaneous, active client connections. The connections are distributed
across the physical links. One client connection will have its traffic
sent on one physical link. The connection is defined by the load-balancing
algorithm. See “Load Balancing” for more information. Each link aggregate can have one or more IP addresses
assigned to it in the /etc/rc.config.d/netconf file. The link partner (the switch, router, or server) ports
connected to the server ports must be configured for link aggregation
(trunking). In addition, the mode on the link partner and the server
must be the same. For example, if ports 1, 2, 3, and 4 are connected
to a link partner switch's ports C1, C2, C3, and C4, respectively,
and the server side is trunked using LACP_AUTO mode, the partner switch
must be configured to trunk ports C1, C2, C3, and C4 using LACP_AUTO
mode. | | | |  | NOTE: MANUAL mode link aggregates using HOT_STANDBY load balancing
can be connected to different switches. In addition, do not enable
trunking on the corresponding switch ports. | | | | |
The link partner (the switch, router,
or server) connected to the link aggregation can inhibit the usefulness
of HP APA in some environments. See “HP APA Configuration Examples” for more information. All the devices in the link aggregation must be the
same type and must be configured for the same speed, duplex, and MTU.
See “Supported LAN Cards” for
the devices HP APA supports.
HP APA provides load
balancing on outbound data transfers using a load distribution algorithm
that you select when you configure a link aggregate. The load distribution
algorithms are based on destination MAC address, IP address, or TCP/UDP
port number. Inbound load balancing is strictly determined by the
link partner (switch, router, or remote server) and has no affect
on the outbound algorithms. Although you can use each of these load distribution
algorithms in all supported configurations, they may not all provide
the same load on each of the physical ports in the link aggregate.
Therefore, HP prefers you use the algorithm that is recommended for
each supported configuration. See “Preparing for Configuration” for more information. The load balancing algorithm consists of the following steps: Data
Flow Lookup — The load distribution algorithm determines
an index into a hash table that includes the physical port through
which the outbound data flow is forwarded. Data
Flow Physical Port Assignment — If the hash index
for the data flow has not been assigned a physical port (the entry
is empty), a physical port in the link aggregate is assigned to that
specific hash index. The physical port is selected on a Round Robin
basis. Aging
Data Flows — Over time, each data flow is checked
to determine if it is still active. If the data flow has not been
active in the last 30 seconds, its specific hash index is cleared
(aged out). If the data flow restarts after being cleared from the
hash table, it is reassigned a new physical port on a Round Robin
basis. For information on clearing a data
flow and initiating the load balancing process, see “Resetting a Data Flow”.
Each load distribution algorithm guarantees that it will
not introduce any severe ordering problems within a specific data
flow. This is required to ensure that the performance is not degraded
significantly as a result of turning on one of the algorithms. Also, all packets for a specific data flow always
flow out through the same physical port until the data flow is aged
out of the distribution table. This means that in order to generate
simultaneous load on each of the physical ports in a link aggregate,
start multiple data flows over the link aggregate. Failover Group | |
HP APA enables you to
combine 2 to 32 physical link ports into one failover group. A failover
group is a link aggregate in LAN_MONITOR mode, but with the following
differences: One port is the active
link, and the others are standby links. Network traffic is sent and
received on the active port. LAN Monitor periodically
exchanges APA packets between the links making up the failover group.
This enables better detection of non-operational links in the failover
group. If the active port or its
link partner fails, LAN Monitor automatically migrates the traffic
to one of the standby ports in the failover group. When a port with
a higher priority than the current active port recovers, the network
traffic is migrated back to the previous active port. Sometimes, it
is desirable to have the network traffic remain on the current active
port after the failure and recovery of the previous active port. To
achieve this, set the HP APA port priorities the same for all ports
in the failover group. You can use 100BT, Gigabit,
or 10 Gigabit Ethernet (10GbE) devices in the failover group. However,
all the devices in the failover group must be of one type: 100BT,
Gigabit, or 10GbE. The failover
group can have one or more IP addresses assigned to it. The physical ports in the failover group do not share a common MAC
address. You can include
link aggregates in a failover group. This enables increased bandwidth
and load balancing in a failover group.
By default, the port in a failover
group with the highest priority is the active port. This is called
priority-based failover. However, HP APA also allows you to configure
failover groups with proactive failover. With proactive failover, the port that is the
most efficient at carrying traffic is the active port. Efficiency
is determined by assigning a cost to each port in a failover group.
This cost is divided by the port's current link speed to yield
a normalized port cost; link speed is the number of links in a link
aggregate multiplied by the speed of a member link, or in the case
of a single link, only the link speed. The lower the normalized port
cost, the higher the link's efficiency. If two links have the
same normalized cost, the one with the higher priority is preferred. For each failover group, if you assign a cost value
to one link, you must assign a cost value to all other links in the
group. If you do not specify a cost value for any of the failover
group's links, the failover group uses the default failover behavior
based on priority. During certain LAN Monitor
events (for example, link failure and link recovery), the normalized
port cost might change on the active or standby links. When these
events occur, the normalized port cost of the active link and the
standby links are compared. If a standby link has a lower normalized
port cost than the active link, the standby link becomes the active
link even if the current active link is UP. TCP Segmentation Offload | |
HP APA supports TCP Segmentation Offload (TSO),
also known as Large Send, on link aggregates and failover groups if
all the Ethernet cards are capable of it. TSO is a mechanism by which
the host stack offloads certain portions of outbound TCP packet processing
to the Network Interface Card (NIC) thereby reducing host CPU utilization.
This functionality can significantly reduce the load on the server
for certain applications which primarily transmit large amounts of
data from the system. In link aggregates, TSO
has the following behavior: If TSO is enabled on all
of the physical ports in a link aggregate, TSO is enabled for the
entire link aggregate. If any of the ports within that link aggregate
go DOWN or UP, the TSO status of the link aggregate does not change.
After the physical ports are added to the aggregate, the TSO capability
of the physical ports cannot be changed. If a port is removed from
a link aggregate, the following occurs: If TSO was supported on
the link aggregate before removing the port, TSO remains enabled on
the link aggregate. If TSO was disabled on the
link aggregate before removing the port, TSO of the link aggregate
is based on remaining ports in the link aggregate. If all remaining
ports support TSO, TSO is enabled on the link aggregate; otherwise,
TSO remains disabled.
If a port is added to a link aggregate, the TSO settings
are recalculated. If the added port has TSO disabled, TSO is disabled
on the link aggregate.
In failover
groups, the TSO status depends on the TSO status of the current active
port. When the active port is changed, the TSO status of the failover
group might change. For example, an active port supports TSO and the
standby port does not. Therefore, the failover group supports TSO.
If the active port goes down, the standby port becomes active and
the failover group now no longer supports TSO. By default, TSO is disabled. To enable TSO on
each specific interface, see the Ethernet Support Guide, available
in http://www.docs.hp.com, in the Networking and Communication section. To verify if TSO is supported on an
link aggregate or failover group, enter the following command: # nwmgr -A vmtu -I instance -S apa
lan900 current values:
Virtual Maximum Transmission Unit = 0 |
In this example, a zero value indicates TSO is not enabled. VLAN Support | |
VLANs over link aggregates and failover groups have the
same advantages of VLANs over physical links, but with the following
additional features:
VLANs over link aggregates offer higher bandwidth
than VLANs over a single physical link. VLANs over failover groups offer improved reliability.
The VLANs continue to carry traffic in case the active link failed. You can use VLANs over one link aggregate to serve
multiple workgroups. This also enables broadcast traffic to be isolated
within the same broadcast domain, offering improved security for workgroups. The same link aggregate or failover group can offer
different level of service for each user using ToS. You gain more
flexibility in how you deploy link aggregates and failover groups. You can create, remove, and modify VLANs over link
aggregates and failover groups without rebooting the system. This
enables you to configure networking on a server without disrupting
other users.
For more information on managing and using VLANs,
see HP-UX VLAN Administrator's Guide and
your switch documentation. Appendix D describes characteristics of using VLANs over link aggregates and
failover groups and guidelines for each configuration. |
Printable version
