812.3ad & Tru64 5.1A PK5

[Arnaud Veron]

Hi,

I'm trying to set up a working 802.3ad link under DS20E Tru64 5.1A PK5 with 2 x DE602, and a cabletron switch ( matrix backplane & 6H202-24 card )
The problem is that it does not work...

my config :

lan_config -i ee1 -a 0 -m utp -s 100 -x 1
lan_config -i ee2 -a 0 -m utp -s 100 -x 1
lan_config -i ee3 -a 0 -m utp -s 100 -x 1
lan_config -i ee4 -a 0 -m utp -s 100 -x 1
lagconfig -c
lagconfig -p ee1 key=1
lagconfig -p ee2 key=1
lagconfig -p ee3 key=1
lagconfig -p ee4 key=1
ifconfig lag0 192.168.0.10 netmask 255.255.255.0 up

Is there something wrong with the above commands ?
The Tru64 kernel is compiled with LAG support. The switch is in 802.3ad mode.

The lag0 interface is running but it seems that only one port is used, and the bandwidth is only about 10 kBytes/s...

Thanks for your help.

[Al Licause]

What is actually happening ? Are you seeing errors or failures ?

What does netstat -i show for ierrs and oerrs ?
Are any of the devices logging errors ?
If so, you may have a speed/duplex mismatch.

Is the switch you are using also configured for 802.3ad ?
LAG is a cooperative between the host and the connecting switch or host.

Which distribution algorithm did you use ?
lagconfig -c dist={dstip|dstmac|port|roundrobin}

Since you did not select one when you created the lag set, it assumes
the default of dstip. Try showing the set to see which set it selected.
This may not be a good match for your configuration.

RE: speed......make sure you have a match between the devices and the switch port.

You usually change the speed/duplex mode on the psuedo device and not
the individual physical NIC's. So lan_config is best applied to lag0.

Check hwmgr to see what speed the devices are currently running:
hwmgr get attr -cat network -a name -a media_speed -a full_duplex

Try removing all but one NIC from the lag set, then add each of the
other three in one at a time. If you have the correct distribution
algorithm, you will see a better distribution of traffic.

Unfortunately, once created, you cannot remove the lag set, but you can
create additional sets using different algorithms. You can move the NIC's
from set to set and see what affect it has.

Also keep in mind that the host algorithm only controls traffic in one
direction. The switch must also set the appropriate algorithm for the
configuration you are using. Check the man pages for more details.

[Michael Schulte zur Sur]

Hi Arnaud,

you have programmed no autonegotiation. Have you set the port of the switch also to fix 100 full? According to man pages the switch also must be configured for link aggregation.

does that help you?

Michael

[Ralf Puchner]

- Construct a link aggregation group out of interfaces that are currently idle. This means the interfaces cannot be marked as "up" in the set up Network Interface Card dialog box of the SysMan Menu, and they cannot have IP addresses assigned to them.

- There has to be two or more of the same type of network interface (Ethernet) dedicated to a single server or switch. The interfaces all be of the same speed and operate in full duplex mode.

- The server or switch that is connected must also be configured for link aggregation.

- LAT cannot run over a link aggregation virtual interface (lag) or any of the interfaces that compose a link aggregation group. So please check if lat module is not loaded during startup of machine or problem may occure!

- Failover is supported on DEGPA (alt) and DE60x (ee) devices only. In addition, the failover time cannot be modified.

Do the following to configure a link aggregation group:

1. Log in as root.
2. Edit the /etc/inet.local file.
3. Enter a lagconfig -c statement to create a link aggregation group.
4. Enter a lagconfig -p statement to enable one port (physical interface) for link aggregation. To enable additional ports, enter
additional lagconfig -p statements.

5. Enter an ifconfig statement to assign an IP address to the link aggregation group virtual interface and enable it.

6. Save the changes and close the file.

7. Restart network services by entering the following command:

# rcinet restart

Be sure interfaces are marked down during configuration.

Please post error messages if problem occurs.

[Arnaud Veron]

Thanks for your very good answers. I tried a point to point lag config between 2 tru64 hosts, each one trunking 2 ee interfaces.
The obtained bandwidth in an ftp session is : 200 Mbps so it is successfull !

The problem is when we came back to the switched config, the switch does not recognize 802.3ad link... bandwidth went down to 100Mbps

I asked network admins to check their switch config.

I hope i can help you in the future !

[Arnaud Veron]

Some (bad) news !

In fact Tru64 support 802.3ad in a certain way... :-)

From HP :
-------------------------------------------
Tru64UNIX does not support 802.3ad (LACP : LAG Control Protocol) over LAG(Link Aggregation).

Tru64UNIX does not reject LACP traffic. But we can not guaranty full bandwidth of LACP over LAG.
-------------------------------------------

The HP contact, by phone, "It might work but it might not work" :-D

[Mario Stargard]

This must have changed. Here's a list of " Multi-port Configurable 10/100BASE-T Ethernet NICs for Use on HP Alpha Systems" which has this in it:

Tru64 UNIX V5.1a IPK includes support for manually configured 10/100 Ethernet Trunk-Groups. The links in a given trunk group must all be the same speed, and must connect point-to-point between the same two systems {(Tru64 AlphaServer-to-(802.3ad compliant switch), or (Tru64 AlphaServer)-to-(Tru64 AlphaServer)}.

From http://h18000.www1.hp.com/products/quickspecs/10462_na/10462_na.HTML

Mario