|
The service is usable for teleconferencing but has annoying glitches. Lately, quality seems to be deteriorating, at least here in central Vermont.
zoomready is open-source shareware I wrote to measure the suitability of an internet connection for teleconferencing. As you can see below, Starlink had an average zoomready rating of 2.66 out of a possible 3.0 over the four measured days. The problem is NOT bandwidth, which has fluctuated but stayed above the minimums needed for good teleconferencing. The problems are failures (most of them short), latency, and jitter. Too often, it takes too long for a packet to get from my machine to the internet and back (latency); the latency varies widely (jitter). Together, subpar latency and jitter make for momentary freezes and poor audio during teleconferencing.
I am now fortunate to have both a Starlink dish and a fiber connection through Stowe Cable. I’ve stuck with Starlink on my machine, but my wife is connected through fiber. Note below how much better the results are when running zoomready on her machine during an overlapping period of time.
There were no failures during the ten days we were monitoring and only a brief period when jitter and latency were subpar, so brief that the average zoomreadiness was 3.0 both for the last hour and for the whole monitoring period.
I’m disappointed in these results. Starlink has improved since I first installed it over a year ago, but lately, it seems to be getting worse here in central Vermont. This decline in service levels may be caused by more users sharing the service; it is still way, way better than traditional satellite, which can’t be used for teleconferencing at all and better than most DSL. It can be used for teleconferencing—I use it that way, but there is a definite quality difference from fiber. The service may improve as more satellites are launched with satellite-to-satellite laser, and other technical improvements are made.
In some rural areas, the latest promises are that fiber is still five years out. Starlink, once you clear the waiting list, is an alternative today. However, unless Starlink improves, its users will be at a substantial disadvantage as the teleconferencing environments of the future require both more bandwidth and lower latency.
If you run Windows and want to monitor the quality of your Internet connection, you can learn more about zoomready and download it free at http://freecheckip.com. It has no ads, does not use cookies, and doesn’t spy on you in any way.
Sponsored byRadix
Sponsored byVerisign
Sponsored byWhoisXML API
Sponsored byVerisign
Sponsored byIPv4.Global
Sponsored byCSC
Sponsored byDNIB.com
Tom,
Check out this demo at an old IETF:
https://www.youtube.com/watch?v=NuHYOu4aAqg
Dave That, who does the demo, has a Starlink list at:
https://lists.bufferbloat.net/listinfo/starlink
and is still interested in bufferbloat.
We tried to do some sort of test for CircleID at my place last week, but could not connect—so I did a post on the roaming failure instead:
https://circleid.com/posts/20220225-spacex-is-testing-starlink-roaming
Here’s a possible scenario. Starlink network must have radio links that connect satellites and access aggregation routers that are located at earth stations. These aggregation links can have variable performance (i.e., throughput and packet loss) due to radio link impairments. During a period of low performance, the aggregation router’s output buffer can fill up due to rate mismatch (i.e., packets are sent to the output queue in the router at rate higher than what the radio link can transmit). This is a hypothesis based on your observations. Starlink engineers need to validate this since only they have access to radio link performance statistics.
I am not a Starlink user. However, I have seen the above scenario in networks that use fixed or mobile radio links.
John Medamana
Actually Starlink is just beginning to deploy satellites which have inter-sat lasers which may be a help. Unlike other LEO networks, almost all of their satellites need to see both a user and a ground station at the same time. I suspect this is part of their problem (downlink congestion) and hope that the new links and possibly orbiting services like DNS services may lead to improvement.
I can see the value of inter-satellite links enabling redundancy using dual/multi-homing of links to ground station. I can’t think of how such links can solve latency and jitter, however. Do you know the motivation behind “orbiting DNS services”? It’s best for DNS architecture to be distributed, but DNS servers can reside at hub locations. I can’t imagine an extra 50ms latency for a DNS lookup to be problematic.
If latency and jitter are partly a result of the hop to the ground station, the more ability the constellation has to distribute that load the better. Orbiting DNS means less need to add hops to and from the groundstation assuming the query can be answered from an orbiting DNS server. I also think frequently used content will be cached first at ground stations then in orbit.