In reply to a post by jchamier:

In reply to a post by Pheasant:

Probs. It's a bit tired to be honest. With the rise of XGS based AltNets there's enough folks out there with 2, 3, 5, 8 Gbps based broadband services.

ISPreview reported the actual takeup of services above 2 Gbps was miniscule in domestic. I know YouFibre is the 8Gbps leader but their coverage is tiny.

Like the rest of this site, it needs to be brought well up to date, as its feels pretty creaky...

Sadly yes, I assume the "day job" where the owners earn money has taken more of their time than before.

Sure, but it will only grow in time. My point is they are behind the curveball here, when you can easily test up to 10 Gbps on off the shelf speedtest.net using various clients etc.

Always strikes me how well Mark does as a one man band over at ISPreview juggling the various demands of news reporting, forum etc

In reply to a post by jchamier:

I'm a fan of wget or curl a super large file and timing it, but that usually is only single connection so in modern WAN networks it appears you need multiple connections to show total throughput.

This is an artefact of the delay-bandwidth product and low level packet loss. The longer the round-trip time to a destination, the harder it is to achieve high TCP throughput.

https://en.wikipedia.org/wiki/TCP_tuning#Packet_loss

e.g. if your destination is 8ms away, and the packet loss is 0.01% (that is, only 1 packet in 10,000 is dropped), then the maximum throughput of a single TCP stream is

1460 / 0.008 / sqrt(0.0001) = 18.25MB/s = 146 Mb/s

This applies to each TCP stream individually, so running multiple concurrent TCP streams increases your total throughput.

Such low-level packet loss can indicate a problem in someone's network somewhere (e.g. dirt on a connector or buffers not sized for microbursts) - but getting your ISP to diagnose such an issue, especially through their upstream transit networks, is pretty much impossible. And it could be a problem at your end, of course.

This sort of optimization *does* happen in R&E networks though, where petabyte datasets are being thrown around. Tools like perfsonar are used to measure the packet loss at very high resolution (by default 10 packets per second = 36,000 packets per hour, measured separately in each direction)

Thank you fascinating analysis!

In reply to a post by Pheasant:

Always strikes me how well Mark does as a one man band over at ISPreview juggling the various demands of news reporting, forum etc

I think the forum here is quite well behaved, I'm not really up on the ISPreview forum, but Mark's news has taken over this place.

Yeah both pretty well behaved. You get the odd weirdo and oddball, but the same anywhere and in any forum I’ve ever used.

I’m glad ISPreview is kept reasonably fresh and up to date.

It’s a slightly different demographic over there too. My impression is a lot of older folks over here, especially the regulars. Perhaps it’s the familiarity of the interface, look and feel. Hasn’t changed in twenty something years. Nothing wrong with that and no disrespect. Just an observation.

In reply to a post by Pheasant:

Nothing wrong with that and no disrespect. Just an observation.

Its the only threaded forum I've used in years, without gimmicks such as "Kudos" that clutter up places such as the official ISP run forums; along with huge avatars. The downside is as Seb reported, its based on an old product with their own modifications. Seb & John have posted they have plans, but I guess time is not easy.