Well, you can get pretty close to this. Check out the Nexland routers: [...]
They can load balance two separate connections. I don't believe this works
for a "single" FTP transfer, for example... it works by loadbalancing
separate transfers. Things like surfing, multi-part transfers, etc. all
happen at 2x the speed.
No, it's a different issue from load-balancing. That's only done when the two connections are both controlled by a single entity, usually at both ends.
The difference is, that under my situation, we are acting as uncoordinated separate entities both on the internet, with essentially a network border between us.
On Wednesday, April 2, 2003, at 01:38 AM, Miguel S�nchez wrote:
While you can think datagram throughput can be boosted this way you may
be surprised that TCP-based applications will be confused about your
setup.
TCP flow control is based on a round-trip time (RTT) estimation for each
connection. While assuming large variations may happen, connections will
work better when RTT does not change a lot (if it does may trigger
unneeded retransmissions that will also reduce congestion window and
will speed down you data flow).
Well there's actually two different situations, (1) my neighbours uses my ISP, or, (2) my neighbour uses a different ISP. To discount situation 2, it's not a problem at all. In situation 2 we don't need to worry about packet reordering as the routes are going to be completely different and IP will probably choose one or the other for any given TCP connection. The problem if the situation is (1) where IP is probably going to be sending packets down the different paths at random since the routes will be identical. So then you have multi-path problems. But, on the other hand, in this case the two paths are likely to be almost identical so that reordering will be minimal. I don't see a problem, the paths will either be very different or very similar.
simon -- general wireless list, a bawug thing <http://www.bawug.org/> [un]subscribe: http://lists.bawug.org/mailman/listinfo/wireless
