On Tue, Feb 3, 2015 at 4:30 AM, Bob Copeland <[email protected]> wrote: > On Tue, Feb 03, 2015 at 02:45:25AM +0530, Ashish Gupta via Devel wrote: >> Regarding the gains, we've also plotted a graph, which you can find >> here, >> https://github.com/scar1337/nodeman/blob/master/experiment/experiment_tcp/ExperimentGraph.png >> That graph should give you some idea about the gains. > > Thanks for the graph! What is time taken in this scenario, time to > transfer a certain amount of data? > > How close are gate 1 and gate 2 spatially? > > -- > Bob Copeland %% http://bobcopeland.com/
We downloaded 16 files of 1M each from the server onto the client. This is done using a python script ( https://github.com/scar1337/nodeman/blob/master/experiment/experiment_tcp/client.py if you are interested) which opens a new socket for each file. Data set for downloading 16 MB of files. All times for Gate1, Gate2 and Combined columns are in seconds. Link speeds were throttled using tc on Linux machines. -------------------------------------------------------------------- Link speed | Gate1 | Gate2 | Combined --------------------------------------------------------------------- 100 kBps link | 169.20 | 183.30 | 91.83 200 kBps link | 92.40 | 92.62 | 47.53 300 kBps link | 61.96 | 62.99 | 31.24 400 kBps link | 46.73 | 47.69 | 23.72 500 kBps link | 37.91 | 37.88 | 19.16 600 kBps link | 32.48 | 32.71 | 21.77 700 kBps link | 27.14 | 27.31 | 17.59 800 kBps link | 24.51 | 25.62 | 17.91 900 kBps link | 24.53 | 30.63 | 16.12 1000 kBps link | 23.98 | 28.72 | 16.01 1100 kBps link | 24.49 | 27.30 | 19.15 1200 kBps link | 24.06 | 26.96 | 16.68 1300 kBps link | 23.84 | 27.02 | 17.11 1400 kBps link | 24.53 | 26.51 | 17.72 -------------------------------------------------------------------- Spatially, the two gates were about 2 metre apart. The experiment was conducted in a somewhat small room so all the machines would have fit in an area of a circle with radius close to 1.5 metre. > Is this similar to link aggregation using bonding > (http://linux-ip.net/html/ether-bonding.html#ether-bonding-aggregation) > that you want to achieve? Regarding channel bonding, it seems like cloud based commercial solutions are available. They talk about speeding up transfers, even when they use a single socket. But there's a catch with that, they use a "speed socket" on their servers to which all our sockets connect and are then routed. Since we are aiming towards a decentralized approach, we don't need central servers running 24/7. As far as I could find, link aggregation is usually performed on a single machine. But here, we are aiming towards more of a "Bittorrent approach for Internet connectivity" so that multiple machines can pool in their bandwidth and all machines in the network can make use of the total pool. Regards, Ashish Gupta _______________________________________________ Devel mailing list [email protected] http://lists.open80211s.org/cgi-bin/mailman/listinfo/devel
