[Bismark-devel] How to measure the initial TCP impulse generated by a web browser?
Dave Taht
d at taht.net
Thu Apr 7 15:26:07 EDT 2011
Jim has been worrying about the size of the initial impulse being
generated by the increase in the initial windows across multiple
operating systems from 4 to 10 or more, and how to look at that brief,
transient impulse, and later, its effect on TCP window scaling.
He messaged me earlier today with:
The best thing I've been able to think is to get a packet capture,
having triggered a bunch of simultaneous HTTP get's on a web server.
Current browsers are doing 6 over the same path (and FF may do 15 under
some circumstances). If you have a sharded web site, you can get yet
more packets in flight, as the web browser doesn't know it's actually
the same web site.
On sites like google images, the connections may be persistent, and the
window size may be growing further when you scroll to the next page. Dunno.
Here's a rough computation
# connections IW Size of "splat" Comments
2 4 8 packets RFC
2068/2616 behaviour
6 4 24 Packets Current
browser + old IW
6 10 60 packets Current
browser + new IW proposal
Not pretty.
Then from timestamp data (even best if it is TCP timestamp data), one
should be able to figure out how long a broadband link is saturated by
the packet burst that arrives.
More information about the Bismark-devel
mailing list