hi Robert,

> On 07 Jan 2015, at 09:13, Robert Kisteleki <[email protected]> wrote:
> 
> Hi,
> 
> On 2015-01-05 14:35, Brian Trammell wrote:
>> Since the target will be an anchor which has known expected content, will
>> these measurements include some indication of whether application-layer
>> rewriting was detected?
> 
> The primary purpose is to measure and to collect and provide this data. We
> can surely implement such checks and related visualisations in the UI and
> perhaps, if there's support for it, in the form of status checks (see
> https://atlas.ripe.net/docs/status-checks/).

That seems reasonable.

> However, first I'd like to see consensus on what data we should collect, if 
> any :)

ACK. :) 

Just connecting to anchors via HTTP gives you a few things: information about 
the routable address of the probe (which I presume is one of the things the 
anchor reply is used for at the moment),  connectivity along a path (but only 
to an anchor) where ICMP is (for some silly reason mumble mumble) impaired. And 
looking at time-to-first-byte and time-to-last-byte changes over time gives you 
some interesting information about the global state of network.

Jared's suggestion would make the whole thing much more operationally useful 
(...says the non-operator...) but at the expense of somewhat more 
infrastructure.

The someone-along-the-path-is-messing-with-HTTP alert I suggested is only of 
operational interest in a few contexts; I mention it because my Big Thing right 
now is trying to quantify all kinds of end-to-end impairment. 

Cheers,

Brian

Reply via email to