Layer 4 data from HTTP Archive?

2 views
Skip to first unread message

Brian Pane

unread,
Jun 30, 2011, 11:53:26 AM6/30/11
to HTTP Archive
Given the existing WebPagetest infrastructure, would it be feasible
for the HTTP Archive to provide tcpdump-type packet traces in addition
to the current page load timeline data?

I have two use cases in mind:
* Estimating network round trip time based on the delay between the
SYNs and ACKs for each connection, in order to better differentiate
network RTT from server processing time in the first-byte-of-response
timings.
* Subtracting the delays due to TCP retransmissions when studying
transmission times.

Thanks,
-Brian

Steve Souders

unread,
Jun 30, 2011, 12:51:20 PM6/30/11
to HTTP Archive
Given that HTTP Archive is only running in one geo location analyzing
the network times is not very valuable because they're from a single
location.

-Steve

Brian Pane

unread,
Jun 30, 2011, 1:38:07 PM6/30/11
to HTTP Archive
On Jun 30, 9:51 am, Steve Souders <stevesouders...@gmail.com> wrote:
> Given that HTTP Archive is only running in one geo location analyzing
> the network times is not very valuable because they're from a single
> location.

I think you might be overlooking an important characteristic of the
use cases I listed: they require only network timings from the exact
same client node that produced the layer 7 measurements.

I'm trying to study a particular class of HTTP responses: those for
which the response time is essentially equal to the network RTT.
Small, static objects often fall into this category. Given the
request timelines from the HTTP archive, if I can identify which
responses are in this "RTT-limited" class, I can then quantify the
effects that things like request pipelining or image spriting would
have if applied to all the sites in the archive. I can attempt to
determine whether any given response in the archive is in the "RTT-
limited" class based on the currently available layer 7 timings, but
having the layer 4 data would enable me to avoid false negatives due
to retransmissions.

Thanks,
-Brian

Steve Souders

unread,
Jul 13, 2011, 3:01:14 PM7/13/11
to HTTP Archive
We don't have that low level timing info.

-Steve
Reply all
Reply to author
Forward
0 new messages