SparkSDR 2.0.4.9

[ahop...@googlemail.com]

Hi Group,

there are some test versions here

http://www.ihopper.org/radio/download.aspx?file=SparkSDR2_0_4_9_win64.zip

http://www.ihopper.org/radio/download.aspx?file=SparkSDR2_0_4_9_win32.zip

http://www.ihopper.org/radio/download.aspx?file=SparkSDR.2.0.4-9.linux-x64.deb

http://www.ihopper.org/radio/download.aspx?file=SparkSDR.2.0.4-9.linux-arm64.deb

http://www.ihopper.org/radio/download.aspx?file=SparkSDR.2.0.4-9.linux-arm.deb

http://www.ihopper.org/radio/download.aspx?file=SparkSDR2_0_4_9_osx.app.tar.gz

These have a faster FT8&4 decoder, just select decode level 0, I't won't decode as many signals as the higher levels but uses very much less cpu.

It also has :-

Delete context menu for profiles and favourites added.

Midi error capture bug that caused program exit fixed.

Bug on computers with > 32 cores fixed.

.deb should now install portaudio on arm and allow for newer versions of libicu.

73 Alan M0NNB

    

[samuel kallmeyer]

Thanks Alan for all these updates,

73 Samuel

--
You received this message because you are subscribed to the Google Groups "SparkSDR" group.
To unsubscribe from this group and stop receiving emails from it, send an email to sparksdr+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/sparksdr/e2beeb65-f559-474d-b565-597b56244cban%40googlegroups.com.

[J P Watters]

Alan,

I loaded SparkSDR 2.0.4.9 on my Raspian PicoCluster Server (5 rPi4's) after I had loaded WSJTX 2.2.2 

SparkSDR does not appear to report any Spots. 

What is the installation process that SparkSDR 2.0.4.9 should follow. 

from the errorlog files

[2021-01-28 22:40:10.754] [sc] [info] set stream id 0 -1

[2021-01-28 22:40:10.755] [sc] [info] pa_start -1

[2021-01-28 22:40:10.761] [sc] [info] set stream id 0 -1

[2021-01-28 22:40:10.766] [sc] [info] set stream id 0 -1

[2021-01-28 22:40:10.771] [sc] [info] set stream id 0 -1

[2021-01-28 22:40:10.864] [sc] [info] set stream id 0 -1

[2021-01-28 22:40:10.869] [sc] [info] set stream id 0 -1

[2021-01-28 22:40:10.873] [sc] [info] set stream id 0 -1

[2021-01-28 22:40:10.877] [sc] [info] set stream id 0 -1

[2021-01-28 22:40:10.882] [sc] [info] set stream id 0 -1

[2021-01-28 22:40:10.886] [sc] [info] set stream id 0 -1

[2021-01-28 22:40:10.890] [sc] [info] set stream id 0 -1

[2021-01-28 22:40:10.895] [sc] [info] set stream id 0 -1

[2021-01-28 22:40:10.899] [sc] [info] set stream id 0 -1

[2021-01-28 22:40:10.903] [sc] [info] set stream id 0 -1

[2021-01-28 22:40:11.199] [sc] [info] set stream id 0 0

[2021-01-28 22:40:11.546] [sc] [info] set stream id 0 0[2021-01-28 22:40:17.269] [sc] [info] set stream id 0 0

[2021-01-28 22:40:17.829] [sc] [info] set stream id 0 0

[2021-01-28 22:40:23.767] [sc] [error] Can't queue seq process

... repeats on infinitum....

Let me know what I am missing.

Meanwhile it is back to running SparkSDR 2.0.4.9 on my Mac Mini

..jpw J P Watters

KC9KKO

Morris, IL USA

[ahop...@googlemail.com]

Hi jpw,

do the basics work i.e does a single receiver running ft8 at level 0 decode anything?  What os are you running on the RPIs? I'm not sure how how all the cluster stuff is meant to work. Are you just installing spark on one of the rpis?  That error generally indicates that you are asking too much of the cpu so how many receivers did you have running?

73 Alan M0NNB

[J P Watters]

Alan,

Yes, I am seeing decodes. Tomorrow night, I will tinker with it to recreate the issue using a single receiver.

I had 10 RX's open as I have with all the other platforms. 

My plan will be to start with a single receiver and go from there. 

As for the 5 cluster of Pi4's, they are all running Buster. 

And yes, just installing on single pi and the cluster software "automagically" shares the load.

Since it appears to be a load issue, I will scale back and methodically see what on receiver does. 

The cluster software is new to me also. I chose to use the "vendors" implementation to start with. 

Steve, KF7O,  is running a different cluster software, I will see what he is doing and try to imitate that.

[J P Watters]

Alan,

My Friday was a tougher day than I planned. I had to work out site in an unheated building installing a fiber run to get our network to a new building. I worked from a lift hanging fiber the rafters of a 35ft ceiling. The helper that I was supposed to have called off sick. So alone again. When I got home late, I ate a reheated dinner and promptly fell asleep in my recliner.

But I woke up after midnight to run SpartSDR 2.0.4.9 on both my MacMini and My Raspian Buster Pi4 Cluster.

With a pair of HL2's I setup a single 80 meter receiver on both machines. The OS X Mac mini showed decodes. The Pi4 did not.

Thoughts?

..jpw J P Watters

KC9KKO

Morris, IL USA

[ahop...@googlemail.com]

Hi jpw,

that sounds like a tough day! My first thoughts of things to try :-

Does level 0 decoding work, if so it would indicate some issue with finding the wsjtx stuff.

Is the clock correct on the pi?

73 Alan M0NNB

[Kristofer Danner]

Alan, confirmed now good with 48 CPU threads. Running 57 virtual receivers at the moment, still using the wsjtx decoder with depth 3. CPUs just loafing.

KD

[Steve Haynal]

Hi Alan,

I'm looking for an arm64.deb file to try on my new odroid-n2+ but these links are stale. Would it be possible to provide an updated linux-arm64.deb?

73,

Steve

kf7o

[ahop...@googlemail.com]

Hi Steve,

I'll build one this weekend, I stopped publishing the arm64 build as I got no feedback on it.

73 Alan M0NNB

[Steve Haynal]

Hi Alan,

Thanks. I look forward to the release. So far I've been pretty impressed with the odroid-n2+. You can see in these benchmarks that it is one of the best performing SBCs, outperforming the Raspberry Pi 4 and the Pine64 or similar rk3399-based systems. The amlogic s922x has 4 performance A73 cores and 2 efficiency A53 cores. I'm hoping the extra cores will help with multiband FT8 skimming. 

73,

Steve

kf7o

[ahop...@googlemail.com]

Hi Steve,

that took longer than it should have, give https://www.sparksdr.com/download/SparkSDR.2.0.8-3.linux-arm64.deb a go.

73 Alan M0NNB

[Steve Haynal]

Hi Alan,

Thanks for the release! So far it has been working fine. I have my odroid-n2+ overclocked to 2.4GHz for the 4 A73 cores and 2.0GHz for the 2 A53 cores. The fan only comes on for a few seconds four times a minute during the FT8 decode phase at the most active times of day. I am skimming all 10 bands at decode level 3. I did build WSJT-X with a different (hopefully better) set of optimization flags. I initially had problems with many EP6 errors but used "cset shield" to assign the network thread to an exclusive A73 core. Now I see one or two errors a minute. I did play with the reserved core setting and found that reserving 2 cores out of the 6 available produced the best results but still needed the cset shield. Performance is better than my J2900 4 core pentium system, but still in the range of a low-end x86. I paid $140 for board, eMMC drive, wifi, case, and fan including shipping. Some of the refurbished x86 thin clients in the $150-$200 range may provide comparable or better performance, but I think the odroid-n2+ will outperform everything that is usually less than $100. The ordoid-n2+ does much better on power. I measure ~3.3W during idle and see it peak to ~6.8W during FT8 decodes.

Do you have any suggestions on how to reduce the EP6 errors other than cset shield? This seems like a very sensitive piece of code. I have increased UDP buffers per this link:

https://medium.com/@CameronSparr/increase-os-udp-buffers-to-improve-performance-51d167bb1360

Since I have 2 cores reserved, I have 4 left for FT8 decode. I see 4 instances of JT9 start. Do all 4 instances have to complete before the next 4 are launched, or will utilization of each CPU be maximized independently? Depending on band conditions or mapping to efficient or performance CPU, one JT9 decode may take substantially longer than others.

The GPU is not being used. This has a Mali-G52 GPU. Does SparkSDR support GPUs on any ARM platforms yet?

73,

Steve

kf7o

[ahop...@googlemail.com]

Hi Steve,

good to hear it is working. That sounds like a good low power setup, I may try one here to leave running on solar power.

I have a few things in the pipeline to help with ep6 errors, one is to make the decoders less sensitive by padding for the missed packets, and there are some perf tweaks still to do on arm.

The ep6 problem appears on all platforms and as far as I can tell is ultimately because all OSs drop packets when things get busy, I do keep revisiting this but have not found a magic bullet yet. 

The decoder threads are independent and grab work when it is available so they do not wait for each other, on a busy setup they can easily be working on different tx time slots. 

The gpu should be being used for the display, if you run it from a command prompt you should see a message indicating if the gpu used, I'd be interested to know what it says.
73 Alan M0NNB

[Anthony N1IG]

Steve,

Curious... are you using the odroid as the remote?  kinda looking for decent options.

Anthony - N1IG

[Steve Haynal]

Hi Anthony,

Yes. I will bolt the odroid to a HL2 in the larger case. I also have an ATU-100 in the larger case. I have LiFE batteries and an easy to setup portable antenna. This is my travel/camping/pota/sota setup. I can setup quickly and then access software running on the odroid via my phone/tablet/chromebook using VNC.

73,

Steve

kf7o

[neiljw...@gmail.com]

Hi Steve,

I have been playing with a Core-I3 machine and finding a fairly high level of EP6 errors.

I think this machine should be similar in performance to the overclocked Odroid-N2+ (Geekbench figures).

I am thinking of getting an Odroid to reduce the power usage of skimming here.

I looked in my notes to remember how to use the cset command, but it seems things have moved on in

SparkSDR and there are now several threads taking a good share of the CPU time. You mentioned shielding

"the network thread" - I wondered how you identified that one?

I tried

ss -up 

to show what was using udp on the network and identified one thread which I shielded, but there are 5 other

threads which are using 4-5X as much CPU. Have I missed anything? I'm getting a similar EP6 rate to you,

around a couple a minute.

Thanks/73,

Neil  G4BRK

[Steve Haynal]

Hi Neil,

I am still using the same formula as described in this post:

https://groups.google.com/g/hermes-lite/c/5nt1t8RpRjY/m/UbHjIGTCBQAJ

I did try shielding more threads, or reserving different processors, but I always saw the best results when shielding just this one thread with its own exclusive processor. Everything else use the other cores. I recently revisited this with a current version of SparkSDR but saw the same results.

73,

Steve

kf7o

[neiljw...@gmail.com]

Hi Steve,

Thanks for confirming the formula and sorry I didn't manage to find the post myself - I did try

but clearly not well enough. I've pasted your description into my notes now.

After leaving it to run for a few minutes the correct thread became clear.

73/Thanks,

Neil  G4BRK

[StefanDausP D]

Hi Steve,

i also have an odroid N2+ here that i'd like to use with SparkSDR, but i'm lacking the time to do the manual setup (especially WSJT-X manual compile etc.). Could you perhaps share a SD card image of your running setup? This would be really great and i know a few OMs that would also love to profit from this.

73

Stefan DC7DS

[Steve Haynal]

Hi Stefan,

I don't want to take on maintaining an odroid image. I do some pretty specialized things, like disable the local GUI when I use VNC and only start it when I run locally, maintain several different netplans, use fixed IPs for my network, etc. It is far from being a setup that is easy for someone else to use.

73,

Steve

kf7o

[StefanDausP D]

Hi Steve,

understood, no problem....perhaps we can at least but together a description of the necessary steps. I'll start to work on that if possible. If you have any notes of your work, I'd be thankful to review them.

best 73

Stefan DC7DS

[Steve Haynal]

Hi Stefan,

I've been very impressed with the stock Ubuntu that comes with the odroid n2+. I purchased the preinstalled emmc drive and it has worked well.

73,

Steve

kf7o

[StefanDausP D]

getting my odroid N2+ ready, now the download link for the arm64 build isn't working anymore...whats the the lastest arm64 version and where can i get it?

BR

Stefan

[ahop...@googlemail.com]

Hi Stefan,

https://www.sparksdr.com/download/SparkSDR.2.0.8-3.linux-arm64.deb where did you find a link that does not work?

73 Alan M0NNB

[Stefan Dambeck]

Thanks Alan,

Link was way back in this thread. 

73

Stefan DC7DS

Gesendet SD mobil 

Am 11.09.2021 um 08:52 schrieb 'ahop...@googlemail.com' via SparkSDR <spar...@googlegroups.com>:

Hi Stefan,

--
You received this message because you are subscribed to a topic in the Google Groups "SparkSDR" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/sparksdr/pc_0YTnQxuE/unsubscribe.
To unsubscribe from this group and all its topics, send an email to sparksdr+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/sparksdr/dfe0c53d-2cd2-41bf-a2df-21a4c3a442adn%40googlegroups.com.

[StefanDausP D]

Hi all,

just to keep you updated, my testsetup using the arm64 build on a odroid N2+ and a red pitaya 125-14 in hpsdr mode is working well. atm i am using 8*96KHz and decoding multiple modes is possible. Like Steve stated, overclocking and active ventilation helps to get the last bit of performance out of this setup.

what a great lean, effective setup this is....thanks again Alan for providing an arm64 version!

vy 73

Stefan DC7DS