Hacker News Reader

Comment by Rangi42 | Hacker News Reader

That's not a practical metric, though. I'm on a campus with a 100Mbit/s connection, I frequently download large files (media and Linux ISOs), and still don't go above 200-300 GB/month. The 100Mbit/s figure is useful for informing me how fast a particular YouTube video will buffer, for example, but by your proposed metric, sustained 100Mbit/s would have to be advertised as 30TB/month. "20 Mbit/s with a 300 GB/month cap" is more informative.

View on HN

I'd like to see the bandwidth measured from my computer to the wider Internet, not from the ISP's speedtest server hosted before the traffic sees the Internet.

At least on my ISP in New York, I can run a speedtest to the auto-selected server and see 330Mbps, but switch to a speedtest server farther away (requiring actual Internet transit), and see the rate drop to 10Mbps. So it's all very impressive that I am paying for a 300Mbps connection to the ISP's colo. But it would be better if I was paying for a 300Mbps connection to the Internet.

(Having a speedtest server hosted by your ISP is a great debugging tool, so I'm not asking them to stop doing that. It's great for figuring out if the bottleneck is my local link layer (WiFi, broken switch?), my Internet connection (broken router? squirrel chewing through a cable?), or simply the ISP's peering choices.)

Looking at these two comments, perhaps we may be reaching an agreement. There are only a relatively few numbers that really matter, although there's tons of complexity. Bandwidth should be a number giving sustained symmetrical throughput over a month + burst capability + download cap, all using the same units. All of those numbers should be 5+ hops away, and should be over a backbone to a server owned by somebody else besides the ISP. In addition, there would be some QoS to show that things like the burst number actually exists for the user 5%+ or more of the time (or some different metric). Heck, there could probably be a bunch more detail you'd add under there, but 3 numbers + access_to_wider net probably covers it.

So in the college example, they're getting (perhaps) .1/100/300

I think if we could just agree on a way to advertise it, we'd be ahead of the game. Net access is becoming like buying PCs: full of jargon and product differentiators meant more to confuse the buyer than assist them.

It's not that impractical. It is how traffic is quoted when renting space in a datacenter.

Reply on news.ycombinator.com