In message <7510E22B-E335-4DA0-8848-514BBE789D19@shagadelic.org>,
Jason Thorpe writes:

>
>On Dec 7, 2005, at 11:10 AM, Nathan J. Williams wrote:
>
>> der Mouse <mouse@Rodents.Montreal.QC.CA> writes:
>>
>>> $DAYJOB has a desire to use multiple network interfaces on a single
>>> machine, with parallel connections to another device, as if they  
>>> were a
>>> single interface with a fatter pipe to the peer device - I believe
>>> cisco calls this "connection bonding".
>>
>> The IEEE term is "link aggregation", specified in 802.3ad. I don't
>> know if this is the same as Cisco's version or not.
>>
>>> Is there any support for this in NetBSD?
>>
>> Nope.
>
>Yes, there is.  See agr(4).

Err..., nope, Jason, you missed a key datum. Yes, we do have agr(4),
but IEEE 802.3ad forbids what der Mouse was actually asking for, as
you  quoted above. My paraphrase of the query:

der Mouse's $DAYJOB wants to spread a (single) TCP connection between
a *single* pair of hosts ("devices"), over multiple links, as if those
multiple links were a [single] fatter pipe to the [i.e., single] peer
device.

AFAIK, agr(4) doesn't support that, , and shouldn't, for the reasons
SMB outlined so clearly.

(OTOH, I do know of switches which can be persuaded to do round-robin
scheduling across links in a link-aggregation group, but that's a
definite case of caveat emptor.)