Re: Logging per-connection Time To First Byte

On 13 May 2011 13:32, Willy Tarreau <w#1wt.eu> wrote:
> On Fri, May 13, 2011 at 12:00:59AM +0100, Jonathan Matthews wrote:
>> Just wondered if there'd ever been any discussion of recording each
>> connection's Time To First Byte in the log output?
>
> Time to end of response headers normally is the same as the time
> to first byte of data, because in order to be able to emit all the
> headers, the application generally needs to have prepared the data.
> In practice, it's extremely common that both the first byte of data
> and the end of headers are even in the same TCP segment. So you can
> consider them equivalent.

Thanks for that, Willy - seems like a very valid point, except perhaps in pathological cases when the response code is hardcoded (or something) hence some headers can be returned significantly sooner than the rest.

Does anyone have any hard stats on Willy's "extremely common" occurrence, above? I can totally believe it's very often the case, but it'd be interesting to get an idea of the frequency with which the entire response header *is* contained in a single packet in general operation ...

TVM,
Jonathan

-- 
Jonathan Matthews
London, UK
http://www.jpluscplusm.com/contact.html