me again. I used www.webpagetest.org to analyse the performance status of a page. After that i did a bunch of optimizations, basically i reduced the http requests on indexpage (which i basically use as “performance indicator”) from 58 to 37, reduced overall datasize by 50 kB ( 10% improvement), put the static files (css, js, some images) to an other hostname (only a subdomain, but tried to configure google analytics to not “track” all subdomains by default (https://developers.google.com/analytics/devguides/collection/gajs/methods/gaJSApiDomainDirectory#gat.GA_Tracker._setDomainName) … i wonder why most people suggest new top level domain, i guess with right GA settings cookie free subdomain is possible, or be i wrong?), and reduced dns lookups (by switching some functions only to load external content if user interact with this functionality, like share buttons etc) …
In fact the page loads way fast now, “everybody” recognizes it when surfs to the page, but the strange thing is: If i test with webpagetest.org i never never never get better results then before optimizations (same settings, server location frankfurt, firefox as browser), that does not make “sense” to me. for example if i test with http://tools.pingdom.com/ ( i know, its chrome and from amsterdam, but server is located near frankfurt … ) results are “always great” - so do you have any hints for me?
if you want to know the actual page im talking about feel free to send a private message!
Try running a test with faster bandwidth and lower latency settings. If your page is bandwidth constrained at the default 1.5Mbps DSL it would explain the difference between pingdom showing gains but WebPagetest not (pingdom tests from the backbone with no latency and high bandwidth).
Since you are using GA, does the site speed data in GA reflect the improvements? That is ultimately what matters since it is the actual performance for your real visitors.
thanks for your answer. To be honest i not especially remember if used advanced settings (more bandwidth, lower latency) for the tests before the optimizations, but in fact i cant except that i have adjusted this advanced settigs for measuring performance before optimizations… :rolleyes:
I have to check the google site speed results in detail, thanks a lot for the hint. but let us discuss, you said “hat is ultimately what matters”.
So i can only measure absolute improvements for the browsers which support “navigation timing”, for the older ones i only can estimate percentual improvments derived from the improvments measured by site speed. Do you can follow me?
Yep. Though you can use the site speed information to break the back-end and front-end performance out (the back-end will be the same for older and newer browsers). You can also look at your regular analytics report to see how much of the user base is on older vs newer browsers.
Did you see an improvement in the GA data when you made the changes? That’s mostly what I was getting at. Since “everybody” says the pages are loading fast, does the data in GA reflect that as well? How closely do the performance numbers in GA line up with what you are getting from WebPagetest? If your user base tends to have much faster connectivity (like they always use your site while at work) then the defaults in WebPagetest may not be representative of your end-users connectivity.
what is different between webpagetest and GA site speed: webpagetest only tests one page, GA tracks all requests and calculates a average load time by default, sure it is possible to view the sitespeed for a specific page, i guess mostly the index page is used with webpagetest… so i have to navigate to the specific site in GA site speed to be able to compare GA site speed with webpagetest.
webpagetest shows definately other results then GA, i checked site speed in GA now in detail and from germany the average load time for the index page in 08/2012 is 1,62 seconds, in webpagetest i rarly get results under 3 seconds…
mhm, i seems as there is no more “connection speed” dimension in GA to check the connection speed for the visitors …
GA measures the performance that your actual users are seeing when they load your site. It is sampled but it effectively has coverage across all of your pages and all of your visitors. You can look at specific pages, specific regions, etc. You will also get a mix of repeat and first page views, etc.
WebPagetest is a single test (or few tests) from a single machine in a single location - basically just one configuration but it provides a lot more visibility into where the time goes.
I’d recommend filtering your GA traffic to:
Then look at the histograms for the distribution and see where the bulk of the visits are. Then go to WebPagetest and try faster connection speeds (Cable would be a good place to start) and see if the times match better.
good point with the filtering for new / returning users, as webpagetest is always a “new visitor” (empty cache).
i now switched from Firefox to IE9 for test settings because GA site speed shows that Internet Explorer is the “fastest” browser for this site (which is also interesting why IE9 “is faster” then chrome or firefox, maybe the ie users have faster connections then the others, mhm?) and uses cable and SEE … no i get a result which is similar to the load speed in GA …
Could have been a UI thing. If you change the browser the connectivity gets reset to the default (an annoyance I’m working on) so if you changed the connectivity first and then the browser you might have run a DSL test unexpectedly.