I did a bulk analysis a few months ago at all of the tests that
had been submitted to WebPagetest here
(24k unique urls): http://www.webpagetest.org/forums/showthread.php?tid=22
One of the stats I looked at was the Time To First Byte (which
as I measure it includes the DNS, initial socket connect and time for the
server to send the first bit of response back).
The knee in the curve for TTFB is around 2 seconds with 90% of
the sites coming back faster than that (76% of the sites came back under 500ms)
– and this is including 50ms * 3 of Round-Trip-Time for last-mile network
latency. 3 seconds would be around the 95th percentile, 4 at
the 96th percentile and 9 seconds at the 98th percentile
(with higher being worse in this case – 98% of sites tested were faster).
If you run a test on WebPagetest
you can pretty much figure out what the RTT to the server is by looking at the
time for the socket connect on the base page (this would be the PING time).
The VAST majority of your time is going to be coming from server resources –
unless your hosting company is in the moon J
Thanks,
-Pat
From:
exceptional-performance@yahoogroups.com
[mailto:exceptional-performance@yahoogroups.com] On Behalf Of GT
Sent: Thursday, November 05, 2009 4:56 PM
To: exceptional-performance@yahoogroups.com
Subject: [exceptional-performance] 'Waiting for server response' and
flush()
Hey y'all
My site's performance has benefited greatly from YSlow and the various hints:
the next step for me is to try to combine images into sprites... I was
staggered to see how much slack there is in loading the underlying images that
are so fundamental to even the most basic website.
FireBug/YSlow has also shown me is that not enough is made of the benefits of a
PHP 'flush()' directive at the first sensible opportunity; this is easy to
overlook but has significant benefits.
One thing that I have noticed though, is that I get significant (>3 or 4
seconds - and up to NINE seconds) "Waiting for server response" on
the initial GET of any page on my server.
The issue is (as I would expect) not as bad on my local clone of the site, but
it's still upwards of 2-3sec and it easily the largest component in the delay
between hitting the URL and getting a 'navigable' page.
It would be terrific if someone could prepare a little guide as to what to
expect as a 'normal' server response time - ow it relates (if at all) to PING
and TRACERT times and how much is determined by server resources. After all, I
can run a debugger and see how long the server takes to 'sticky-tape' the
actual page together... and it's nowhere NEAR 3 seconds (it's more like 0.3) -
so it's not an issue of the server making the browser wait 7 seconds while it
preps the page.
I tried using teh.Google (gasp!) to get some answers, but no joy.
Cheerio
GT
Australia