Re: [Bloat] facebook releases network test tool for slow and dodgy conditions

[Jonathan Morton <chromatix99@gmail.com>]

> On 24 Mar, 2015, at 21:57, Steinar H. Gunderson <sgunderson@bigfoot.com> wrote:
> 
> On Tue, Mar 24, 2015 at 03:49:42PM -0400, Rich Brown wrote:
>> I hope they *do* have bufferbloat in their emulator. It'd be a real shame
>> if they failed to mimic the presence of bufferbloat in 3G/4G/DSL/etc.
>> uplinks. Imagine if everyone developed their app using ATC, then wondered
>> why their performance stinks in the real world...
> 
> Does the distinction between “low-throughput, low-latency” and
> “low-throughput, low-latency, bufferbloated” really mean all that much for
> the average userspace app? You have to go pretty deep in the stack before the
> _reason_ for the slowness matters much.

I’d say that there’s a useful distinction between “low throughput, *high* latency” and “low throughput, low latency, bufferbloat”.  The distinction between the two cases you mention should be obvious to everyone here, of course.

The distinction between general high latency and bufferbloat-induced latency is that you can potentially mitigate the latter at the application level, by opening fewer connections or performing manual congestion control.  This assumes that you’re the only application using the link, but it’s better than nothing.

 - Jonathan Morton