Hi Manuel, Quoting Manuel Bouyer 30/07/2013 11:58,
load-balance is intended to be used for direct link between 2 hosts supporting this mode. AFAIK a switch is not capable of doing load-balance.
Still, normally it's possible to double the bandwidth with LACP, although it may be necessary to have two connections to prove it (download or upload two ISO images at the same time, for example, with FTP). So even with the default mode (no load balancing), we should get the better bandwidth. I haven't tested that.
Any of those is not supposed to work with non manageable switches, right?in the general case, yes. load-balance could work in some specific cases with non manageable switches, but you'll get the extra bandwith only for transmit, not receive.
It is nice to know that possibility.
I tested the default -link0 (dash) mode between two netbsd nodes, and it didn't really work out of the box, I had to manually recreate the interface on one of the nodes to establish the link and ping, even though the '0x3<COLLECTING,DISTRIBUTING>' came right up. Also once the link seemed to be alright, when I physically disconnected *one* of the wires, I lost the connection. So I wasn't very successful with this, could be my mistake though, I should try again. I didn't try to validate the link0 (w/o the dash) the same way, but it seemed to me that I also had to reinitialize the agr0 interface on one of the two nodes to make it work.what network interface do you use ? I found that in most case you have to change the hardware address of ethernet interfaces before creating agr(4). All ethernet interface part of the same agr(4) have to use the same ethernet address.
Those first testings were made with two re(4) cards on one side, one alc(4) and one re(4) card on the other. Both being NetBSD hosts.
Hmm I don't clearly remember if the MAC address were changed automatically by the agr configuration.
On the other hand, as for the link1 mode, I am happily using it on my home network on two netbsd hosts, both connected to a non manageable switches, and everybody -- even the nodes with no aggregates -- pings.but then you don't get the extra bandwith.
In the end I figured out I had terrible performance with this mode on normal switches. I just disabled agr for the time being and used motherboard's alc(4) interface. Either there's hardware acceleration that makes it work marvelously again compared to agr(4), although the box on this side is pretty much powerful enough to deal with any kind of load, or there's an issue with agr(4) link1 mode. Does NetBSD support hardware acceleration? I doubt it -- Anyway I've got great performance with alc(4) alone compared to an link1 mode aggregate with the alc(4) and re(4) interfaces.