Softether maintenance - got a spare image?

Mike va3mw · January 2017

All

I have been pounding away on HF with my Maestro and my RPI running Softether almost a year now.

About a month or 2 ago, I noticed that I could not write to the file system on the RPI connected to the back of the Maestro. I really noticed it when the Softether Server manager had issues keeping its settings.

It became apparent that the uSD card was going bad. I wasn't too worried as the Maestro continued to function 'well enough'.

Tonight, I built a new image and got it online. In the last hour, I have dropped ZERO packets when I would have dropped a bunch more. I am looking at 0.00% dropped over a 240,000 packets from my remote base 150km's away. I can even now up the frame rate without breakup. (It did help that I also did a lot of bufferbloat tuning at the remote end where all I have is 2MB/s upload).

While I was at it, I hard coded an 'fsck' on boot each and every time. I added '/usr/bin/touch /forcefsck' to /etc/rc.local so that the Pi always does a forced fsck of the root filesystem on each boot.

Now that I have a new image up and running, I am going to: a) make a copy of the image and b) make a spare uSD card that I will attach in the case of the Maestro 'just in case'.

Yes, I should have done this sooner.

Mike va3mw

Dave Taht · January 2017

I am curious as to if you enabled ecn during your bufferbloat tuning. Because otherwise congestion control is designed to drop packets.

Mike va3mw · January 2017

Dave

I am not sure what ECN is, so I am afraid I can't answer that. However, I do have QOS turned and have limited my uploads to 80% of my perceived maximum (80% of 2MB/s)

I also reduced the transmit buffers in DD-WRT down to about 25 from the default as this is V2 or DD-WRT. I will replace that router with one that supports CODEL in the spring when I get up to the remote base. I don't see it from October until about April.

Here, at the receiving end, I am running a pfSense router, but I am still tuning it.

This is after having the Maestro run all night on 80M JT65. Not one packet dropped of my Softether Network.

Mike va3mw

Dave Taht · January 2017

ECN - explicit congestion notification - is a way of doing lossless congestion control that fq_codel supports (assuming your tcps have it on). There is no other way to do congestion control without some loss, so when you said you didn't have any - and were happy with that! - I assumed your stats were wrong or you'd enabled ecn.

From what I see above, I suspect you are measuring things at the interface, after congestion control happened elsewhere. It's great you have such good connectivity...

but your max latency (2.5 seconds) is horrific.

So if you deploy fq_codel on the pfsense router and the other side, you will see some loss on the path (if measured at the right place) but dramatic reductions of latency under load to about 10-20ms over your baseline.

I hope whatever you end up deploying ends up with code like this on it, on the wifi chip itself, which works better than everything we've deployed before.

https://lwn.net/Articles/705884/

Mike va3mw · January 2017

Hi Dave

Thanks for that. There is no Wifi, this is hard wired all the way through. The 2500ms latency was a 1 time event which I say happen, but cannot explain why. 99% of the time I am well under 100ms. Thanks for the link BTW. Very informative.

Mike

Softether maintenance - got a spare image?

Comments

Leave a Comment

Howdy, Stranger!

Categories