Per-Entity Metadata Working Group

[Tom Scavo <>]

On Tue, Sep 6, 2016 at 2:17 PM, Patrick Radtke
<>
 wrote:
>> Okay, testing my theory...the diff (time_starttransfer -
>> time_pretransfer) in each case is:
>>
>> 0.211
>> 0.090
>> 0.085
>> 0.086
>> 0.086
>>
>> The first one is clearly an outlier...any idea why?
>
> I'm not sure. Repeating the test now, I get a diff of
> 0.195
> 0.086
> 0.170
> 0.085
> and then ~0.085 for the next 10 in a row.
>
> It might just be the variability of last mile connections.

From my understanding of the curl timings, that doesn't follow. The
difference (time_starttransfer - time_pretransfer) captures the time
interval from the end of the HTTP handshake to the very beginning of
the data transfer. In other words, that time interval is completely on
the server (so to speak) so a caching server with no load should give
consistent timings (with little variability).

I'll keep testing...

Tom