AutoPort Agregation Not Balancing

[Lolowsuper]

Hi everyone,

we´ve been checking our link agregation and we´ve checked for all the configurations in 3 machines that we have with this configuration. One of them is pretty fine, but the other 2 we´ve just checked with our network staff that for only one interface the packages are passing by... the other one is not loading anything or only a few things.... does anyone have any clue about this?

Regards,

Flavio.

[Tim Nelson]

Post your hp_apaconf, looking to see what your loadbalance method is set to.

[Lolowsuper]

Hi Tim,

HP_APA_INTERFACE_NAME[0]=lan900
HP_APA_LOAD_BALANCE_MODE[0]=LB_CPU
HP_APA_GROUP_CAPABILITY[0]=1
HP_APA_KEY[0]=1
HP_APA_MANUAL_LA[0]=5,7

HP_APA_MAX_LINKAGGS=5

###########################################################################
# The HP_APA_INIT_ARGS are reserved by HP. They are NOT user changable.
HP_APA_START_LA_PPA=900
HP_APA_DEFAULT_PORT_MODE=MANUAL

HP_APA_INIT_ARGS="HP_APA_LOAD_BALANCE_MODE HP_APA_GROUP_CAPABILITY HP_APA_HOT_STANDBY HP_APA_MANUAL_LA HP_APA_INIT HP_APA_KEY"

# End of hp_apaconf configuration file

[Lolowsuper]

The 3 machines have the same configuration...
but only one machine is balancing!!!!

here is the lanadmin statistics:

Enter Port PPA number: 0

**** PORT NUMBER: 0 *******
Port Mode : MANUAL
Port State : PROTOCOL OFF
Port Group Capability : 5
Port Priority : 0
Link Aggregation specific test mode. Valid LAN Interface PPAs: 5 7

linkagg = Link Aggregate status
port = Port's APA status
end = End Driver Specific Test Mode, return to Lan Interface
test mode.
menu = Display this menu
quit = Terminate the Administration, return to shell

Enter command: linkagg

Link Aggregate PPA # : 900
Number of Ports : 2
Ports PPA : 5 7
Link Aggregation State : LINKAGG MANUAL
Group Capability : 1
Load Balance Mode : CPU Based (LB_CPU)

Link Aggregation specific test mode. Valid LAN Interface PPAs: 5 7

linkagg = Link Aggregate status
port = Port's APA status
end = End Driver Specific Test Mode, return to Lan Interface
test mode.
menu = Display this menu
quit = Terminate the Administration, return to shell

[Lolowsuper]

why am i saying this? we can see at glance too that it´s not balanced... That link agregation interface is used by oracle rac...

here:

the balanced one:

6 lan5 Lan 1947.2/1690.0 2241.8/ 1809.0 841.8/ 817.7 1118.1/ 972.6
7 lan6 Lan 0.0/ 0.0 0.0/ 0.0 0.0/ 0.0 0.0/ 0.0
8 lan7 Lan 1881.8/1623.6 1869.0/ 1809.7 942.7/ 763.8 948.1/ 963.1
9 lan900 Lan 3829.0/3313.6 4110.9/ 3618.7 1783.6/1581.6 2065.4/1935.7


the other 2 machines:
6 lan5 Lan 1813.0/3596.4 3101.0/ 3513.0 851.0/1893.1 1227.0/1681.2
7 lan6 Lan 0.0/ 0.0 0.0/ 0.0 0.0/ 0.0 0.0/ 0.0
8 lan7 Lan 0.0/ 0.0 0.0/ 0.0 0.0/ 0.0 0.0/ 0.0
9 lan900 Lan 1813.0/3596.4 3101.0/ 3513.0 851.0/1893.1 1227.0/1681.2


--------------

6 lan5 Lan 4869.0/3869.7 2313.0/ 3578.4 2401.0/2088.7 1274.0/1775.1
7 lan6 Lan 0.0/ 0.0 0.0/ 0.0 0.0/ 0.0 0.0/ 0.0
8 lan7 Lan 0.0/ 0.0 0.0/ 0.0 0.0/ 0.0 0.0/ 0.0
9 lan900 Lan 4869.0/3869.7 2313.0/ 3578.4 2401.0/2088.7 1274.0/1775.1

[Dave Hutton]

Are the switch ports set correctly to do it?

[Tim Nelson]

Not that this fixes anything but LB_CPU is obsoleted in 11iv2. This is according the the sept 2006 apa admin guide.

load_balance_algorithm LB_MAC (default), LB_IP, LB_PORT, LB_CPU (LB_CPU is obsolete in 11i v2)

[Lolowsuper]

Well,

i´m using HP-UX v11.11... and for one machine it works fine.. we´re assuming that network staff did configured the switches as the same for the 3 machines... and they do have the same hardware ...

[Lolowsuper]

do you know if is possible to rediscover an interface in the link agregation, or check for the status in software... or so... anyway... a method that we can check for the possible problem that might be ocurring in this network interfaces.

[Lolowsuper]

in another post Dave Olker said something about a WTEC tool for link-monitor, doesn´t anyone has that?

thank you..

sorry for the bad english.

[Tim Nelson]

There is a pile of info and even vendor switch setup stuff in the below admin doc.

http://docs.hp.com/en/J4240-90039/J4240-90039.pdf

lanscan -q
lanadmin
netfmt
are a few commands to check stats and status.

[Tim Nelson]

One other item to note: We are also asumming that all three servers have the same LAN driver versions, patches and APA software versions installed ??

Another easy test.
from server1 ftp to server2 and transer a large file
at the same time
from server2 ftp to server1 and ftp another large file

Review your interface stats to see if both physical interfaces in the aggregate are sending data.

I have seen specifically in an 11iv1 and LB_CPU that outbound traffic will latch onto a single nic and inbound to another.

[Lolowsuper]

Ok Tim,

I´ve doble checked for patches, drivers and APA version and it´s the same on the 3 machines.

The test you suggested i won´t do it because I´ve already many connections that oracle rac generates, so lets assume that we have a load average on it!!! as you can check in a few replys back that i´ve posted a glance cut.

What i would like to try in the first instance is if is possibile to restart the APA process as listed in in the Admin document. Could i make it online? like if I do so am i going to shutdown the lan900 interface and i will lost all conections, or if I do so, the service will not shutdown the interface but will try to reload any kernel modules or something like it...

[Tim Nelson]

If you bounce the aggregate you WILL loose LAN connectivity over LAN900

You might be able to manually drop 1 leg and re-add it but I would definately do some testing in a non-prod envrionment first. If not available then wait for maintenance time to test.

[Lolowsuper]

Tim,

what do you mean in manually drop one leg? Using software or physically unplug the cable?

[Tim Nelson]

Either one.

You can disconnect a physical cable or use lanadmin commands to manually add or remove a nic from the aggregate.

[rick jones]

If you are using the "CPU" balancing then the balancing of traffic will depend on how the scheduler balances the processes/threads generating traffic. If all the threads generating traffic happen to be run on the same CPU for whatever reason, you won't see any spread.

That of course is all for outbound traffic. For inbound traffic it will all be up to the switch.

[LucianoCarvalho]

Hi,

Please post the switch configuration. If it's a cisco product:
Show interfaces summary
show interface port-channel

I think that you could try to change the load balance algorithm.


#lanadmin -l -X -l load_balance_algorithm linkAggregatePPA

Sets the load-balancing algorithm for a link aggregate with the specified
linkAggregatePPA number.
The following choices are valid for load_balance_algorithm:
â ¢ LB_MAC
â ¢ LB_IP
â ¢ LB_PORT
For a description of these parameters, see the load distribution algorithms in â Link Aggregate
Advanced Parametersâ (page 37).

best regards