Hi Ashish, Thank you for sharing your interesting work with the list. Are you relying on open80211s to discover the gates (i.e. using gate announcements)? Or do you use some other mechanism in user space?
Best, Javier On Sun, Feb 1, 2015 at 10:01 AM, Ashish Gupta via Devel < [email protected]> wrote: > Hi! > > This is regarding multiple gates in o11s. My team and I found the idea > interesting and decided to explore it at a greater length. We have used the > Click modular router to set up an SDN with multiple gates in the mesh > network. We've conducted a few experiments and the gains using two gates > seem to be additive. > > We were wondering if this would be beneficial and feasible for the current > implementation of o11s as the gains seem to be promising. There also seem > to be valid use cases as multiple gateways in close proximity are pretty > common in apartment societies, to name a place. Community networks may also > be a good ground for such deployments. If there is a way this could make it > into the kernel, we are more than willing to put in the required effort. > All of your suggestions are welcome and I am sure they can help us make the > most out of this project. > > We have effectively moved away from single associations , which is the > convention, towards a model of multiple associations, where each node in > the mesh can be "connected" to several internet access gates at the same > time. This is being done on a per flow basis so that we can route all the > traffic associated with a single socket through one gateway. This is done > in order to not confuse a server on the remote end as it might otherwise > treat packets from a different gate as a DoS attempt. > > We've been experimenting with a few types of gates such as an Android > phone tethered via USB, which effectively replicates a USB dongle, and also > wired ethernet connections. We've also conducted a few experiments using > o11s as the local network with Click scripts running on each machine. The > experiments were attempts at replicating the real Internet by using a local > http server and a local Bittorrent tracker. > > Details of the local experiment follow with an HTTP webserver (nginx) : > > We host a single 1 M file on the server (S) which is in turn accessible > by two machines (G1 and G2, the two "gates"). The client (C) wants to > download multiple copies of this 1M file from the server, > using these two paths : > > C -> G1 -> S > C -> G2 -> S > > Each of these paths resemble a single internet connection. The aim of this > experiment was to study how the network reacts when we try to use both > connections (paths) at the same time for higher throughput. > > We varied the link speeds in the scenario and plotted a graph for the link > speeds versus the time it took to download files using a single G1 path, a > single G2 path and both paths respectively. > > The number of simultaneous file downloads was set to 16 and the link > speeds were varied from 100 kbps to 1400 kbps using the tc utility on both > the intermediate machines (G1 and G2). > > The gains are linear till a point where the hardware link seems to > saturate. > > This is the public repository on github : > https://github.com/scar1337/nodeman > > It has a README and setup scripts which will help you set up and hack away > at the project in case you are interested. You will need our modified > version of click which is available at https://github.com/scar1337/click > > Needless to say, all of the machines do this using a single interface > instead of multiple interfaces that most research papers talk about. Thanks > for reading the long mail. I hope this can be extended somehow so that o11s > networks can use all available bandwidth and experience maximized > throughput. > > Regards, > Ashish Gupta > > > _______________________________________________ > Devel mailing list > [email protected] > http://lists.open80211s.org/cgi-bin/mailman/listinfo/devel > >
_______________________________________________ Devel mailing list [email protected] http://lists.open80211s.org/cgi-bin/mailman/listinfo/devel
