HN Explorer

Ask HN: Why Is Speedtest Upload Test Disrupting Pings?

2 points by adir1 ↗ HN
We have been recently blessed with Gigabit Fiber connection from AT&T. In my unscientific attempt to gauge connection stability I usually run 100 DNS pings and SpeedTest - all via latest Win 10 connected directly to router by short Cat 6 (gigabit) ethernet cable.

Today I noticed serious slowdowns in Ping but ONLY during Upload portion of the Speedtest, which itself reported 700mb/s. While 940 mb/s Download test has almost no impact on ping timing. Any idea what may explain this?

Reply from 8.8.8.8: bytes=32 time=3ms TTL=53 Reply from 8.8.8.8: bytes=32 time=4ms TTL=53 Reply from 8.8.8.8: bytes=32 time=4ms TTL=53 Reply from 8.8.8.8: bytes=32 time=5ms TTL=53 Reply from 8.8.8.8: bytes=32 time=5ms TTL=53 Reply from 8.8.8.8: bytes=32 time=4ms TTL=53 Reply from 8.8.8.8: bytes=32 time=6ms TTL=53 Reply from 8.8.8.8: bytes=32 time=3ms TTL=53 Reply from 8.8.8.8: bytes=32 time=3ms TTL=53 Reply from 8.8.8.8: bytes=32 time=82ms TTL=53 Reply from 8.8.8.8: bytes=32 time=165ms TTL=53 Reply from 8.8.8.8: bytes=32 time=222ms TTL=53 Reply from 8.8.8.8: bytes=32 time=266ms TTL=53 Reply from 8.8.8.8: bytes=32 time=285ms TTL=53 Reply from 8.8.8.8: bytes=32 time=303ms TTL=53 Reply from 8.8.8.8: bytes=32 time=321ms TTL=53 Reply from 8.8.8.8: bytes=32 time=337ms TTL=53 Reply from 8.8.8.8: bytes=32 time=352ms TTL=53 Reply from 8.8.8.8: bytes=32 time=368ms TTL=53 Reply from 8.8.8.8: bytes=32 time=379ms TTL=53 Reply from 8.8.8.8: bytes=32 time=389ms TTL=53 Reply from 8.8.8.8: bytes=32 time=282ms TTL=53 Reply from 8.8.8.8: bytes=32 time=3ms TTL=53 Reply from 8.8.8.8: bytes=32 time=3ms TTL=53 Reply from 8.8.8.8: bytes=32 time=3ms TTL=53 Reply from 8.8.8.8: bytes=32 time=3ms TTL=53

It starts around same time as speedtest started, and at the end you can see "normal" results when speedtest is finished. You can see that download test reported higher speed and had barely any impact on ping test timings.

3 comments

[ 3.2 ms ] story [ 23.3 ms ] thread
pwg
The most likely cause is bufferbloat: https://www.bufferbloat.net/projects/bloat/wiki/Introduction...
adir1
Interesting point, I have read about it in the past - mostly in the context of cable networks though.
pwg
That may be simply the result of those who were initially investigating the effect having cable internet connections.

But there is nothing about bufferbloat that is exclusive to "cable internet". The effect will happen anywhere that a too large buffer exists in a device with differing speed input and output links that tries too hard to never, ever, drop a TCP packet.

The effect you describe, ping latency skyrockets during upload phase of the speedtest, is exactly what happens when some device in the path of your packets and pings has an overlarge buffer that has filled. The most likely culprit is the device installed by AT&T that interfaces between your local network and their fiber network (i.e., the fiber "modem" or "router" box).