Upload success rates - vary with package and time of day

3 posts / 0 new
Last post
G3ZIL
Offline
Last seen: 1 week 1 day ago
Joined: 2015/06/08 - 07:15
Upload success rates - vary with package and time of day

While working with Rob, AI6VN, and Phil, VK7JJ on making sure that the WsprDaemon package was uploading all of the spots it decoded I saw that my own upload success rate for WSJT-X 2.4 could be below 90%. This led to a more in-depth study with the headline findings listed below.

This is a different issue to topic http://www.wsprnet.org/drupal/node/8776 started by Jim, WA2ZKD, on spots that had arrived being dropped by wsprnet.org - this topic is on spots that never arrive.

I don't have enough knowledge to know exactly what's the root cause, and whether the success rates could be improved by changes at wsprnet.org or within the WSJT-X package. Hence I'm posting to both forums. Any insights from this community would be most welcome. There is a detailed report - I could not attach as a file, I reported the error, so here is a Dropbox link.

https://www.dropbox.com/s/pihb3ljy6aa0jse/Upload_success_rate.pdf?dl=0

Summary of findings
1. The upload success rate with WsprDaemon is 100%. Spots are uploaded in bulk using MEPT (curl to http://wsprnet.org/meptspots.php).

2. The upload success rate using KiwiSDR WSPR 1.3 extension is about 99.9%. Most of these spots arrive well after those from WSJT-X and WsprDaemon within a two-minute interval.

3. The upload success rate for WSJT-X depends on type of computer used, affected by the load on the wsprnet.org server, and varies with time of day. Success rates for a Pi4B running a single instance of WSJT-X range from 88% to 97%. Success rate for Mac Mini, Macbook pro and an HP i5 quad-core 1.8 GHz laptop with Windows10 varied from 99.5% to 99.9%.

4. Examination of TCP transactions for failed spot uploads show instances of no receipt of acknowledgements from wsprnet.org yet POSTs from the WSJT-X client and wsprnet.org resetting the connection. WSJT-X itself seems not to retry failed uploads.

5. It's not easy to draw a conclusion from an analysis with Spotnum as a proxy for time of uploads within a two-minute interval. In particular we cannot reconcile the relative 'tightness' of the Pi4 spots' Spotnums, suggesting the spot-by-spot uploads were not drawn out in time, with the fact that only 9 out of 24 were uploaded in that interval.

Gwyn G3ZIL