Link Aggregation failure

Do I need to change any configuration settings in the switch level

Absolutely. In order for link aggregation to work, both the server and the switch need to know what to expect. Turning it on at one end of the link is going to have problems - right in line with what you're seeing.

You will need to configure the switch to use link aggregation on the ports your server is connected to, as well as configure it on the server itself.

Well, I have the EXACT same problem. MacPro 2.66 with Leopard Server 10.5.1 and LevelOne GSW-2494.

I have created a bond interface on the OS, and set the ports on the switch as LACP. The two ethernet cards are switching between red and green lights. Ive also notice that when I copy a file to the server, the procedure "hangs" for seconds when the lights are red.

The GSW-2494 says in the manual that is 802.3ad capable.

P.S. If i break the NIC bond, and connect one card on the switch, everything is fast and with no problem.

Thanks

Kostas

MacPro 2.66 quad with Leopard Client 10.5.1.
Dell PowerConnect 2724. (Supports 802.3ad according to Dell. Reports of success with WinXP.)

Set the two relevant ports on the Dell as a LAG group and set up a Bonded interface on the Mac. Reset the switch and when the switch comes back up the interface works for about 90 seconds. After which the Mac reports that the switch isn't configured for 802.3ad.

This is a consistent scenario. Works for 90 seconds after the links are brought up and then Leopard drops the ball.

Well, I gave up. By setting up the LACP in both the MacPro (10.5 Server) and the Level1 switch, I got almost half the speed in File Services. The Level1 representative claims that the switch is 802.3ad capable and tested with Windows Servers NICs, so the problem is in the Apple box.

I will try to setup a Tiger Server in a MacPro and check if it is a Leopard-specific problem.

I will post back anything new.

Kostas

Not that this helps in your situation, but I setup Link Aggregation on our new Xserve to a Cisco Catalyst 4006. Haven't had an issue, I've tested copying files and dropping a port without the copying stopping.

Hopefully you find a solution.

I am using a DLink DGS 3100-24 and cannot get it working. 10.5.1, Xserve Xeon. In 10.4.11 on other equipment this config would work for me. Not sure what's at Fault, 10.5.1 or my ability, or the D Link.

I'm going for what's probably an "unsupported" configuration in OSX by trying to use lacp with each port on an xserve connected to a different switch (cisco 3750). The 2 switches are trunked together and I assign the same port-channel # on each switch. The config looks like this:

switch 1:
interface GigabitEthernet1/0/14
channel-group 16 mode passive
switchport access vlan 211

switch 2:
interface GigabitEthernet1/0/41
channel-group 16 mode passive
switchport access vlan 211

What happens is only 1 switch will activate the port-channel interface. The other port shows link up but line protocol down - with no active lacp packets. this jives with what OSX shows me - claiming one of the ports is OK but the other "port is connected to a partner but the link aggregation group is different from the one that is active". I even managed to get the lom to follow the active port by configuring the same IP on both ethernet devices in the server monitor configuration (since this configuration doesn't seem to know about the bonded interface).

When I shutdown the active port, the traffic switches over to the other port/switch (by way of the port-channel interface switching ports/switches). this is mostly what I want ... however it takes about 30 seconds for this to happen. I think this is because OSX is only requesting slow LACPDU's on the active port according to the switch:

sh lacp 16 neighbor
Flags: S - Device is requesting Slow LACPDUs
F - Device is requesting Fast LACPDUs
A - Device is in Active mode P - Device is in Passive mode

Channel group 16 neighbors

Partner's information:

LACP port Oper Port Port
Port Flags Priority Dev ID Age Key Number State
Gi1/0/41 SA 32768 001e.52f1.fb84 926s 0x1 0x5 0x34


So my question is, how can OSX be configured to request Fast LACPDUs instead? Any ideas?

I'm going for what's probably an "unsupported" configuration in OSX by trying to use lacp with each port on an xserve connected to a different switch

This is completely unsupported and won't work. The host (in this case, the server) has to negotiate with the switch and they have to agree on which links are in the trunk. Since each switch can only 'see' one link from the server it will assume that the trunk is down and that's as far as you'll get.

You might be able to connect the server to two stacked switches, but regular switches trunked together is never going to work.

That's mostly the response I was expecting. I've got a similar problem in Solaris, while in Linux it works just fine. Ignoring the separate switches, how about getting the OSX server to use fast rate lacp instead of slow rate?