Re: Proposal to measure end-user latency

I love both. Out of bound measure is not simple in all cases ( per example with strong auth and complex PKI, for health application, ...).
What and how to measure?
Of course, DNS lookup, connect time, time to first byte, ... are interesting but it may use a lot of bandwidth. 
So, let say that people who need all that do it at application layer (like W3C specify it), and simply measure the total elapse time from DNS (if any) to last byte. 

Le 4 sept. 2013 à 18:19, Mike Belshe <mike@belshe.com> a écrit :

> I love the idea of metrics.  I think the only thing I love more is simplicity.  
> 
> I think the choices for measuring latency out-of-band from the protocol itself are numerous - so this ends up just being more stuff.
> 
> The debate about what-to-measure, how-to-measure, etc would be very lengthy. 
> 
> Mike
> 
> 
> 
> On Wed, Sep 4, 2013 at 8:45 AM, Roberto Peon <grmocg@gmail.com> wrote:
> I suppose the question here then is how much better these (client-side) measurements are as compared to what the server can observe and generate on its own, and what attack vectors this information might possibly allow/enhance.
> 
> On Sep 4, 2013 5:46 AM, "Sébastien BARNOUD" <sebastien.barnoud@prologism.fr> wrote:
> My proposal wasn't to have an interaction between the application layer
> and the protocol. So, I also agree.
> 
> My proposal is to have an optional measurement at the protocol level
> itself. Uses cases could be for protocols over HTTP, like SOAP to have
> automatically a minimal set of measurements without extra coding at
> application layer.
> Of course, it will be available for a HTML application that doesn't
> implement any measurement at the application layer which, although it is
> regrettable, is often the case.
> 
> However, it is true that a protocol is not there to overcome all the
> misery of the world.
> 
> 
> Le 04/09/13 07:00, « Willy Tarreau » <w@1wt.eu> a écrit :
> 
> >On Tue, Sep 03, 2013 at 03:56:49PM -0700, Roberto Peon wrote:
> >> I'd imagine that you'd want to re-use the same timing information that
> >>is
> >> collectable within the javascript, and likely in the same format (that
> >>just
> >> makes sense from an engineering perspective).
> >> That format and when each of the timestamps is collected is currently
> >> defined in the W3C.
> >>
> >> I'd just ask them to trigger the collection of that data collection upon
> >> the condition stated in the header.
> >> The interaction with HTTP would just be to add that header to the IANA
> >> registry, if so... doesn't seem very likely to require this WG.
> >
> >If it'd be done that way, I totally agree.
> >
> >Cheers,
> >Willy
> >
> 
> 
>