[Bloat] Latency Measurements in Speed Test suites (was: DOCSIS 3+ recommendation?)

Dave Taht dave.taht at gmail.com
Mon Mar 30 10:55:03 EDT 2015


I think the most effective thing would be to add bufferbloat testing
infrastructure to the web browsers themselves. There are already
plenty of tools for measuring web performance (whyslow for firefox,
the successor to chrome web page benchmarker) more or less built in...
measuring actual network performance under load is not much of a
reach.

The issues this would resolve are:

1) speed - the test(s) could use native apis within the browser and
thus achieve higher rates of speed than is possible with javascript
(and monitor cpu usage)
2) we could rigorously define the tests to have similar features to
netperf-wrapper
3) we could get much better tcp statistics as in with TCP_INFO
4) output formats could still be json as we do today, but plotted better
5) ?

Problems are:

0) Convincing users to use (and believe) them
1) Suitable server targets for the tests themselves
2) Although the browsers are basically in a nearly quarterly update
cycle, it would still take time for the tests to be widely available
even if they were ready today
3) Convincing the browser makers that they could use such tests
4) Writing the tests (in C and C++)
5) The outcry at speedtest, et al, for obsoleting their tools
(think microsoft vs "stacker")
6) Bloating up the browsers still further



More information about the Bloat mailing list