You should be seeing performance increases with multi mode trunking. The default behavior is to hash the MAC address and assign all MACs within a specific range to a single port. you can also do it by IP address. The upshot is that you won't see a performance increase if you're testing from a single client, but when you have a couple of dozen hosts you should see the packets load-balanced across all the ports in the vif.
There's an option to do it round-robin style, which theoretically would result in higher performance from a single client. However, the delay and processor overhead in re-ordering packets usually negates (or at least significantly reduces) the benefits of trunking.
--paul
On Fri, 25 Mar 2005 12:25:00 -0800, Chris Moll CMMoll@lbl.gov wrote:
Has anyone gotten any performance gains from network interface trunking with a netapp? So far I haven't.... my suspicion is that I don't have the switch setup right.
More details:
I'm trying to setup trunking on a netapp connected to a Cisco 3550 gigabit switch (OS vers.12 on the switch).
The odd thing is that when I setup two gigabit lines trunked together from the netapp to the switch I saw no improvement in thoughtput compared to a single gigabit line.
As a test I set up four 100Mbit interfaces trunked together from the netapp to the switch. Again, I see the same throughput as with a single 100Mbit interface alone - there was no gain in throughput from adding three more interfaces. (Measured by connecting a Sun to the switch with a gigabit interface).
The trunking worked even before I set anything on the switch. The only thing I knew to do on the switch was put the interfaces in a channel group:
switch> interface range GigabitEthernet1/0/17 - 18 switch> channel-group 1 mode on switch> end
...but still no change in throughput (this worked with sun trunking, and there I did see a gain in throughput).
netapp> vif create multi romvif0 -b rr e4a e4b e4c e4d
netstat and vif status both show roughly equal numbers of packets on all four interfaces; all four are definitely being used.
-Chris Moll Lawrence Berkeley National Laboratory