[mosh-devel] multipath mosh

[Dave Taht]

Groovy. You are based on the new release?
On Aug 1, 2015 5:54 AM, "Matthieu Boutier" <
boutier at pps.univ-paris-diderot.fr> wrote:

> Hi all,
>
> First thanks to all developers and maintainers for the 1.2.5 release.
>
> I also have the pleasure to announce an update for my multipath version.
> The main changes are:
>
>   - link-local addresses now works well,
>
>   - add loss ratio computation for each path, and use it to limit the
>     global loss ratio,
>
>   - simplification of the sending procedure.
>
> The incoming loss ratio compution is based on the path sequence number
> with a 64 bits sliding window, and the outgoing loss ratio is received
> from the remote peer.  For this purpose, I took 8 bits of the flag field
> in the network protocol.
>
> The paths are still ordered and selected by RTT, but multiple paths can
> be used to send the same packet (duplicates), to achieve a configurable
> maximum loss ratio (the -m option), which defaults to 0.  If you set it
> up to 10(%), and if your 3 best (RTT) paths have an estimated loss ratio
> of respectively 30, 20 and 0%, then only the two firsts will be used,
> having together a loss ratio estimated to 30% * 20% = 6% < 10%.
>
> Remark that paths with a loss ratio estimated to 100% are used, except
> if they are idle for too long.  This allow us to set the loss ratio to
> 100% on new paths: it's particularly usefull at the beginning, where all
> paths are used together.  Idle flows are quickly discarded.
>
> The source code is at:
>
>     https://github.com/boutier/mosh
>
> Enjoy,
>
> Matthieu
>
> _______________________________________________
> mosh-devel mailing list
> mosh-devel at mit.edu
> http://mailman.mit.edu/mailman/listinfo/mosh-devel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.mit.edu/pipermail/mosh-devel/attachments/20150801/daa3a9be/attachment.html