[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: Extending shim6 to use multiple locator pairs simultaneously
On 03/21/06 at 10:08am -0800, Tony Li <firstname.lastname@example.org> wrote:
> Here's another thought that's worth what you paid for it:
> Instead, create a single TCP connection for each link that you want to
> use. Create a muxing layer on top of these connections that distributes
> data across the connections in a possibly uneven manner, filling the
> buffers of each connection. This layer would require some additional
> framing and sequence numbering. On the receiver side, demux using the
> framing and then reorder the arrived data. Now, you're using the full
> bandwidth of all of the available links, and saturating them up to their
> full TCP performance level.
Sounds like BitTorrent. :)
The problem I see with this is in creating the multiple TCP connections
for the multiple locator pairs. Without some sort of signalling (aka
shim6 context establishment) the two hosts can't know each others'
locators. Without that knowledge, you don't know if you're talking to
multiple locators on the same host, or multiple different hosts proving
the same service (if you get the multiple IPs via DNS, for example).
So I see the incorporation with shim6 as a better fit than trying to
re-create many of shim6's features at layer 4.