This is really just a success story about how a stock install of Gargoyle 1.5.5 with QOS and ACC turned on is doing an exemplary job of arbitrating amongst four greedy users on a 3Mbit connection.
Users:
Projector laptop, running uTorrent wide-open,
227 TCP sessions and 514 UDP connections
A friend's laptop as a "seed box" running uTorrent wide-open for upload and download
74 TCP sessions and 256 UDP sessions
Office laptop, streaming 128Kbit audio, using Gmail and various remote-control sessions
47 TCP sessions and 14 UDP sessions
Nexus 7 Android tablet running international video-skype session
33 TCP sessions and 5 UDP sessions
A few other Android tablets with ten or so sessions apiece.
Connection: cable modem, 3 megabit download, 256 kilobit upload.
This is not a big, fast, luxurious connection. This is an economical connection and so efficiency is critical.
Connection is completely maxed out; flat-lining at the top of the range.
Well of course it is; there are not one but *two* torrent clients on the network, either one of which would take all of the bandwidth and trash the latency and responsiveness of the network.
Yet the international music streaming is proceeding without a skip, browsing is fluid and snappy, and when I asked about the quality of the int'l Skype video call the answer came back "great, why?"
The connection is jam-packed, right? The Bandwidth Graph shows the line is completely full. One would expect poor speedtest results under such conditions - the line is already jam-packed.
I just ran a speedtest and got 2.86Mbit and 150Kbit upload. Gargoyle instantly "made room" for this high-priority traffic.
I got 95% of the download speed I would have gotten if the line were empty.
And I also got 60% of my upload.
And latency on the Speedtest.net speedtest came back at 22ms which is amazing given all of the heavy traffic of all kinds from many users.
Watching the Bandwidth Usage graph, I can see a tiny "chasm" where Gargoyle made room for the speed test, and then it took a moment for the torrenting laptops to re-take the slack.
A glance at the Active Congestion Control section shows that the MinRTT traffic class is active. Well that's the Skype call, right? I've chosen to optimize WAN throughput rather than minimize ping times, and my pings are still 60ms range even without MinRTT active.
Anyway, Gargoyle with stock settings, and Active Congestion Control turned on is doing this amazing job.
It really has me thinking about 200-user deployments...
Gargoyle doing an amazing job
Moderator: Moderators
Re: Gargoyle doing an amazing job
exellent
Re: Gargoyle doing an amazing job
A little bit of follow-up on testing...
In subsequent "retests;" meaning hitting the "TEST AGAIN" button on Speedtest.net, I got lower values.
Different values corresponded to different cases.
Worst case was two Youtubers and two torrenters, and simply hitting "TEST AGAIN"; speeds here were 20% category.
With just two torrenters, hitting "TEST AGAIN" would yield 70% of linespeed.
Actually reloading the page and starting the test again fresh made for 80% or better speedtest numbers with two torrenters on the network.
Last test case was running a YouTube session on this machine, then loading the speedtest page and testing during the YouTube session, which was still loading the video. Remember there are still two other torrenting machines active on the network also. This test yielded an amazing 98% of linespeed.
So Gargoyle was able to completely throttle back two torrenting machines, and selectively throttle back the YouTube streaming on this machine to favor the speedtest running in a different tab on the same browser. Delivering 98% of linespeed.
On stock settings - amazing.
Does this behavior square with the theory or would you have predicted different outcomes?
In subsequent "retests;" meaning hitting the "TEST AGAIN" button on Speedtest.net, I got lower values.
Different values corresponded to different cases.
Worst case was two Youtubers and two torrenters, and simply hitting "TEST AGAIN"; speeds here were 20% category.
With just two torrenters, hitting "TEST AGAIN" would yield 70% of linespeed.
Actually reloading the page and starting the test again fresh made for 80% or better speedtest numbers with two torrenters on the network.
Last test case was running a YouTube session on this machine, then loading the speedtest page and testing during the YouTube session, which was still loading the video. Remember there are still two other torrenting machines active on the network also. This test yielded an amazing 98% of linespeed.
So Gargoyle was able to completely throttle back two torrenting machines, and selectively throttle back the YouTube streaming on this machine to favor the speedtest running in a different tab on the same browser. Delivering 98% of linespeed.
On stock settings - amazing.
Does this behavior square with the theory or would you have predicted different outcomes?