Ed said:
He has a point.
Consider dual CPUs and how they interact; and then think of bit-torrent
downloads. What the OP's suggesting is certainly feasible, but as to
whether or not there's available software to do it, well, I don't know.
Ed
Some of the terminology can be found on this page.
http://smallvoid.com/article/winnt-network-load.html
BitTorrent may be able to take advantage of what is described in the
first part of that article. If a torrent uses multiple connections,
when each connection is set up, it may be possible to randomize the
connections over the two interfaces. For the BitTorrent protocol,
this could result in a higher transfer rate.
It would be harder to get a protocol like FTP, to work with that,
because when the (single) connection is set up, it'll all be going
over one interface. So granularity is an issue with some of
these schemes, and BitTorrent has the solution for that. (Yes, there
are FTP methods that open multiple connections, so I suppose you might
argue that FTP isn't totally out of the question. You could be downloading
a different file on each link, as a trivial example.)
They also mention "Teaming", but teaming seems to involve two
cables running to the same piece of networking equipment. So
I'm not sure that would do the same kind of thing.
http://en.wikipedia.org/wiki/Link_aggregation
http://en.wikipedia.org/wiki/Channel_bonding
"Multiple dial-up links over POTS can be channel-bonded together
in the same manner and can come closer to achieving their
aggregate bandwidth than routing schemes which simply load-balance
outgoing network connections over the links."
On ADSL, there is actually an option at the terminating equipment called
MLPPP. My ISP offers this, for $4 per month over base ADSL pricing.
And some user on one of the web forums, has actually tested bonding
together seven ADSL connections (they all have to terminate at the
same ISP, and the same terminating equipment). So seven links was
the maximum the equipment at the other end would support. Apparently
MLPPP can even be run, as a protocol, on one link, with the intent
being, to prevent other "filtering" equipment, from throttling
certain protocols a customer might use. So the option is even useful
on one link (until the filtering equipment has been updated to
consider this situation). I don't know if Cable/DOCSIS has
an option like that or not.
http://en.wikipedia.org/wiki/MLPPP#PPP_Configuration_Options
You can also find forums which discuss stuff like this.
http://www.broadbandreports.com/forum/r25899747-Cable-vs.-2-Link-MLPPP-Comparison
Two cable modems, could end up connected to the same segment of
shared cable. At least, the original implementation of a cable
setup, would look like this.
Head End ------- coax_down_your_street ----+---------+---------+
| | |
cable cable cab;e
modem modem modem
Now, depending on the subscription level, or speeds involved, a city
street could be chopped up into shorter cable segments. But it wouldn't
be as likely, that one household would have physical access to two
separate cable segments. What I'm trying to get at there, is there is
the possibility of resource contention, so "bonding" the devices
isn't necessarily a win. For example, during the busy hour, if you
see your bandwidth plummet, bonding might double the bandwidth for you,
but the doubling might not even reach the level of what a single link
could achieve during the non-busy times. So if the intention was to
achieve a "guaranteed minimum bandwidth", even two cable modems
might not be enough, if the cable segment is over-subscribed.
You might see 20Mbit/sec on each link during the quiet hour,
and 3Mbit/sec during the busy hour, and 2*3Mbit/sec still isn't
a lot (compared to the 20Mbit/sec during the good times). If you
were having trouble watching IPTV on a single cable connection,
then if the segment is congested, perhaps two bonded together
wouldn't be enough. For a protocol like BitTorrent, you wouldn't
care about that, because it's the average rate that counts (over
a longer period of time).
All Internet networking options, work on the premise of concentration
and over-subscription. That takes advantage of the customers that
are "quiet" most of the time, to stretch the equipment further.
Even modem pools in dialup days did that - you could easily subscribe
to dialup, go to use it, and get a busy signal. Over-subscription
in broadband, leads to peak bandwidth reduction during the busy hour, so
at least some service is still available.
I never notice a reduction in my ADSL service, but then, it's so slow
compared to cable, who cares... :-( My download rate is a steady 300KB/sec.
Big deal. Seven times that, can easily be beaten by some of the newer
cable or FIOS offerings.
For cable, DOCSIS 3 is the latest popular standard.
http://en.wikipedia.org/wiki/DOCSIS
"DOCSIS 3.0 features channel bonding, which enables multiple downstream
and upstream channels to be used together at the same time by a single
subscriber.
Common DOCSIS 3.0 speeds are listed in the table below.
downstream upstream Downstream Upstream
channels channels throughput throughput
8 4 343.04 Mbit/s 122.88 Mbit/s "
So if a provider had a monthly offering of that nature, even a single
modem connection could give hefty performance. And that is bonding
of channels available on the shared cable. It might not be necessary
to bond at the cable modem level, if channel bonding can give the
desired result. I suppose the difference might be, if you had two
different cable segments, but that isn't likely to happen.
Paul
