Starlink: Outage Data End of June Update

Published Friday, July 2, 2021 by Bryan

June was an exciting month for networking in the Northwoods, with lots of new data collection activity.

June 2021 connectivity timeseries. 1sec per pixel, 20min per line. Red=obstruction-downtime, blue=other-downtime, green=no-satellite, dark-grey=no-data (reboot)

The bright red covering the top end of the graph shows how bad the obstruction downtime was for the first ten days. June 10 got all the way up to 70min per 12hr!

Then the db6b5d22 firmware installed on June 11, and wow, what a difference! We saw obstruction downtime drop below 30min per 12hr for the first time in over a month … maybe two.

Outage length histogram, June 1-10, 2021.
Outage length histogram, June 11-18, 2021.

The histogram of outage lengths may give some clue to what changed. Notice the spikes at 15sec and 26sec in the June 1-10 dataset, and their absence (or shift to 11sec) in the June 11-18 dataset. I think we may finally be seeing the effects of the alternate-connection change the Starlink beta participants about in April. If that's the case, it works, and it's a big improvement, because I haven't moved my dish since my last update, but my connectivity has definitely improved. (Stats stayed similar after the 19th, when the a4908729 firmware installed.)

The new obstruction view included with the new firmware.

The other fun change in the new firmware is the new obstruction view in the app. The very low detail wedge diagram has been replaced with a silhouette of the sky. This is what I was aiming for when I was trying to "raster-scan" the scene back in February. This will be very helpful for my next relocation or tree-trimming. I can already tell that one of the three trees I was most concerned about is not a problem!

I've mentioned before that while this is the "better than nothing beta", I'm not actually comparing Starlink to nothing. I also have a fixed-wireless ("WISP") connection that is operated by a local company. This month I set up my own monitoring for that connection as well.

Ping success of my fixed-wireless connection in June 2021. All outages in blue, because it's not a beta service. The regular dark-grey diagonal is an artifact of my sampling setup.

This is both what allows me to delay resolving my Starlink obstructions, and what initially pushed me to join the beta in the first place. Look at all of that regular connectivity (white/clear) … and that large band of unconnectivity (blue). It's a generally stable connection, unless it goes down after business hours, and then it's down until business hours the next day.

For the first ten days of June, I often used the fixed-wireless connection instead of Starlink. When obstruction/other outages get bad, sometimes lower speed with higher connectivity is less frustrating. But how much lower is the speed? I set up monitoring for that as well.

Fixed-wireless speedtest stats, grouped by hour of day. X-axis labels are 'Hour (Successes count / Test count)'

I pay for the maximum speed my WISP will give me: 25Mbps down, 5Mbps up. Looking at the stats, the only time I average near that is between midnight and 8am (the left three box-and-whiskers are from 12:50am, 3:30am, and 6:10am). During the workday, download speed sometimes reaches that level, but averages more in the 15-20Mbps range (the next four plots are 8:50am, 11:30am, 2:10pm, and 4:50pm). Through the evening, the average dips below 15Mbps, and even as low as 7Mbps (the right two plots are 7:30pm and 10:10pm). So, it's usable during the day, but things can get a little slow in the evenings.

Starlink speedtest stats, grouped by hour of day.

My setup runs a speedtest on Starlink immediately after running one on my fixed-wireless. The variation is far greater - Starlink isn't limiting me to a purchased bandwidth cap. But notice the shift in scale: that lowest grid block was 5Mbps for the WISP, but is 50Mbps for Starlink. Only one sampling bucket, 2:10pm, has a minimum download speed that is obviously lower than 25Mbps, the fastest my WISP produced. There is also little to no apparent evening dropoff. This isn't improved speed - this is a different class of speed.

Fixed-wireless speedtests, grouped by day of week.

I will at least give my fixed wireless provider props for apparently managing bandwidth well enough that differences between weekend/weekday are not apparent. I live in an area that sees its population double in tourist season, with weekends being the biggest influx. Neighbors warned us that networks might be more congested with more people around, but so far that seems to mostly not be noticeable.

Two final bits of speedtest trivia:

  1. Over 270 tests, Starlink used approximately twice as many Ookla servers as my fixed wireless provider (20 vs. 11). I presume this is because I connected to a server that was well connected to the ground station I was connected to at the time, and my ground station changed.
  2. Only five of the servers were used by speedtests on both my WISP and on Starlink (for 26 total unique servers).

So that's June. It sounds like mid-August is supposed to be when the constellation is able to provide full global coverage. That makes me look forward to the connectivity improvements we may see in July as more satellites get into place. This speedtest data has also nudged me farther toward getting an arborist out here.

Categories: Starlink