TOR is often too slow, this happens when one of the nodes chosen at a time is too slow.

Instead we could imagine using several TOR connections simultaneously and redundantly, ignoring the packets that take too long to be acknowledged so only the fastest connection takes the full load.

How can this be done?

Maybe with virtual machines acting as TOR clients and presenting a network interface each to the host so the host thinks it has 10 network interfaces and is duplicating its packets across them?

Maybe something similar to EOP Xtream, but for linux and open-source and with the connections to different network interfaces gatewayed to TOR clients?

http://www.brianmadden.com/blogs/vid...-goodness.aspx

I'm not certain you wouldn't achieve the same efficiency by bonding a pair of NICs and running multi-path.
You're looking for the shortest route between two points, let's hope re-inventing the compass is optional.

Any good places to read about and get started with multi-path?

Here's a place to start

Red Hat has DM Multi-path capability, but that's specifically for storage, not server client arrangements.

i do think that would make you VERY identifiably in the tor network
and rather easily tracked back to YOU

negating the use for TOR

just use a normal network for SPEED

What would visited sites see? Wouldn't they see 10 visitors all downloading the same page?

Multiple mostly useless requests for the same page in a short time from the same site, across multiple paths of a finite network (TOR)...

That could be construed as a form of DDOS attack in my book!

At best it is wasteful of the target site and TOR network resources, very selfish and inconsiderate! At worst, with multiple clients it would become an anonymized DDOS delivery network in its own right! At worst, if it became the defacto way to use TOR it would leverage the traffic levels by the number of request paths - 10X in your example! A 10X level of traffic on an already slow resource limited network... you see where that goes...

What if multipath only uses one of the paths, the one with the fastest acknowledgements?

1 members found this post helpful.

Just found it does something close to that. Here is a demo video (see under "Linux Kernel MultiPath TCP in action"):

http://www.multipath-tcp.org/

I don't think IP multi-path sends identical requests over multiple paths, it polls for a faster path than it currently routes over and when it locates one, it uses it.
So concerns about it creating accidental ddos attacks are probably misplaced.