Hacker News Reader

Comment by deepsun | Hacker News Reader

SpeedTest.net tests are useless, as ISPs give higher priority to traffic to known speed checkers.

They can legally do that as we don't have Net Neutrality anymore.

Eh kinda... priority traffic shows the maximum potential speed of your link. The max speed of your link has nothing to do with the speed you'll get on any particular node on the internet. For example if the ISP has too little peering bandwidth to some other speed test.net host you will still be capped by that link.

These days I rarely need large bandwidth to any single host, but have a lot of bandwidth use between many hosts including streaming applications. Bufferbloat is far more apt to be problematic under this workload than ISP priority, unless there is a bandwidth shortage on the local links.

I believe that's why Netflix introduced the fast.com speedtest, which runs through Netflix servers. Reducing the likelihood that the test gets shaped.

The link - https://fast.com

Cloudflare's - https://speed.cloudflare.com

fast.com is the only browser based speedtest that shows the accurate results for me. The results from there resembles the download speeds I get when downloading files from different sites or installing upgrades through my package manager.

Cloudflare's speedtest often performs worse for me.

iperf3 - https://iperf.fr Is the most accurate CLI speed test I've found (depending on which server you choose).

What's "accurate" is going to be different for every website / destination network anyways. Unless you suspect there's some issue with the speedtest software running on the browser / target, it's showing you the accurate speed you get to Cloudflare or whatever server you're trying to pull from. If you just want to see what your ISP is limiting your bandwidth to, sure, you need to find a server that can hit that.

The best speed test is to download some large linux ISOs with 100+ peers, so you max out your connection to tons of networks.

"accurate" is for me the speed I have on average throughout the day no matter what endpoint and not the max througput. If I'm doing a speedtest, most of the time I just want to know if I'm now downloading file $x from $y what download duration can I expect.

Downloading packages with $PACKAGE_MANAGER, downloading files from $CLOUD_STORAGE, downloading videos from youtube, downloading linux ISOs, receiving files through $INSTANT_MESSENGER, ... they all are very close to the speed I get with fast.com.

That's the reason fast.com comes from the same servers as Netflix content - it's difficult to prioritize one without prioritizing the other.

Is this true? Do you have a link?

Not coming in with proof but I have personally seen this- a buddy and I did some testing and the Ookla tests were the only ones we did that were much higher than the others.

I'm sorry for not contributing with hard evidence but I've seen this behavior (though I never heard about it from anyone else before now)

I used to work for a small ISP. The boss specifically wanted speed tests to be exempted from traffic shaping.

Not surprised. I have long noted that when my my browser is strugeling to load tabs all i have to do is try connecting to speedtest.net for all open tabs to suddenly load and i have decent speed for the next 15-30 minutes.

https://speed.cloudflare.com/

I like this one because your speed really depends on what you're downloading and this one tests some of those scenarios.

Reply on news.ycombinator.com