Gargoyle Forum

I've been keeping an eye on how ACC works, and over the past few weeks noticed the following.

Background info: I have my download QoS on with just one category and let IP sharing take care of equally splitting bandwidth between devices. I have a 5/1 plan, but I usually get around 5.7 Mbps download.

What I've noticed: When I saturate the line (I get two computers to download Ubuntu simultaneously for testing) my fair link limit goes to 5900kbps and that's when pings shoot up to like 800ms. They stay high enough until the fair link limit goes down to 4900kbps. And the loop goes again. Slowly it goes up to 5900, with pings staying real low until then, then 800ms and collapses down to 4900kbps again. Every minute or two, the fair link limi goes from 5900 to 4900 to 5900 to 4900 to... and so on. Right now on auto, the ping limit is set to 49ms, but I've also tried with MinRTT (28ms) and a custom 150ms ping limit. I've tried changing the IP to Google's and OpenDNS servers, and it's the same loop. Last night I tested it with two Ubuntu downloads for half an hour... to see if it would eventually stop, but it kept doing it.

I guess what I don't understand is, why is the observed ping really low (under 20s) from 4900 to 5800, but when it hits 5900, ping goes up to 800ms and stays in the hundreds long enough to make the line limit drop to 4900kbps. If full load stops when the line limit is at 4900, my connection will stay capped at 4900 until someone places full load on it (watches an HD video or downloads something). It seems like ACC is too aggressive in quickly lowering the line limit, or not noticing ping decreases quickly enough.

Just wondering if there's anything I can try to prevent this or if this is normal behavior. Other than that, QoS is working great. Thanks.

I'm running 1.6.1 on a Linksys WRT400N.

Your experience is not totally unexpected to me. When your ISPs pings go from 50ms to 800ms it does cause the ACC to react quickly and the type of oscillation you describe is what I have seen in my testing as well. Although with my 1Mbps line the swings are less dramatic.

I have done my best to make the ACC behave as well as I can but I do not have access to every setup in the world or the time to optimize each if I could. To change the response time of the ACC will require you to examine the source code and change the gain of the digital controller there, recompile and test. You are welcome to try and see if you can get a better result. I tried to make the code clear so you can see how it works. Good luck and report any results you obtain.