[LINK] ACS new? submission on ISP filtering

[Crispin Harris]

This time I send it to the LIST rather than just Stilgherrian :_)


On Fri, Mar 19, 2010 at 9:29 AM, Stilgherrian <stil at stilgherrian.com> wrote:

> On 19/03/2010, at 11:38 AM, Roger Clarke wrote:
> > I suspect that "triples the latency" is a serious
> > over-estimate.  No, I don't have a simulator handy to experiment
> > with, but presumably someone has).
>
> Don't need an experiment, Roger, it's east to demonstrate. And I'm about to
> prove myself wrong by showing it's only double, not triple the latency.
>
> For any content not hosted in the next room, the major cause of latency
> would be the travel time for the packets from client to server and back.
>
> Asking for content with HTTP: Trip 1 HTTP GET request sent. Trip 2 Data
> returned.
>
> Checking with HTTP HEAD first: Trip 1 HTTP HEAD request sent. Trip 2 Data
> returned. Trip 3 HTTP GET sent. Trip 4 Data returned.
>
> Close enough?
>
> Either way, the lag between requesting content an it starting to appear on
> screen is the biggest factor in users deciding to skip off the site. I don't
> see slowing this process down being much of an advantage.
>
> Then again, why are we even discussing this? Isn't worrying about this what
> we have network engineer minions for?
>
> As one of those much-maligned engineering types maybe this is time for me
to throw in 2c ?

The major difference will come down to implementation of the session.
Each time an application BEGINS negotiation of a TCP-based session, a number
of things have to happen:
[Three way handshake]
  Q: Anyone There?
  A: Yep, you there too?
  Q: Yep! Lets chat!
[Data Session]
  Q: HTTP GET <blah>
  A: HTTP DATA <blah>
  Q: Thanks, got Blah1
  A: OK, here is <blah2> <blah3> etc...
  Q: Thanks, got Blah2, Blah3 etc...
[Finishing/cleanup]
  A: Finished now, Bye
  Q: Excellent, me too. Bye.

Now if you have to do this twice, the connection initiation latency can
really add up.
Lets assume that you are getting your packets from somewhere inside
australia, and thus your Round-Trip Time is less than 150ms. (if you are
really lucky, and live in Sydney, your RTT may be 70ms or less - but we will
run this simulation at 150ms).

With a relatievely optimised session and no additional levels of back/forth
the session above can transfer a 3.5kb no-image web page in aprox
600ms (to completion - about 300ms to initial data availability).

MINIMUM decision time is RTT*2 (and this assumes that in-host turnaround is
instantaneous, and that you don't have protocol-stack or application
delays).

If the process of HTTP HEAD, followed by HTTP GET closes a session and then
opens a NEW session for the HTTP GET request, then the THEORETICAL MINIMUM
time to (App-receive data) is RTT*4.

If the response on the HTTP HEAD and/or HTTP GET is longer that 1300 bytes
of content, then add a minimum of 1 instance of RTT for each.

Also note: you cannot being retreiving any images until AFTER you have
retreived the initial HTML page.

As you can see from this discussion, **LATENCY** (round trip time) is the
key determinant in page-load performance.
and HTTP 300 redirect, any framesets and (non-serverside) includes
substantially increase page load times because of the number of new session
negotiations involved.

If your RTT latency gets up over 300ms you can really feel the delay in page
load.

If the HTTP HEAD and HTTP GET is streamed in the same session, the times
decrease somewhat.

<grin>

C


> Stil
>
>
> --
> Stilgherrian http://stilgherrian.com/
> Internet, IT and Media Consulting, Sydney, Australia
> mobile +61 407 623 600
> fax +61 2 8569 2006
> Twitter: stilgherrian
> Skype: stilgherrian
> ABN 25 231 641 421
>
>
>
> _______________________________________________
> Link mailing list
> Link at mailman.anu.edu.au
> http://mailman.anu.edu.au/mailman/listinfo/link
>



-- 
Crispin Harris
crispin.harris at gmail.com
"A great deal of Security is unfortunately just like the underwear of
Brittany Spears.
If it's even there at all, it is needlessly complex and frilly; looks good
without actually covering much; and is far to easy to get around or remove
completely."
- David Boston

Marriage (n): a natural institution whereby a man and a woman give
themselves to each other for life in an exclusive sexual relationship that
is open to procreation.
-Definition compliements of Cardinal George Pell, Catholic Archdioces of
Synd