I just remembered a problem I came across for visual comparisons - I can’t compare first run with second run (with primed cache) of the same page.
My use-case is obvious - to see how caching affects the speed of rendering.
It would probably be also meaningful to allow to compare two cached versions of different pages (e.g. dev vs. production) too.
I hope it’s easy to do - it looks like images are already collected (filmstrips are different for them) and it’s just a matter of providing the user interface.
Well, there’s “can’t” and there’s “can’t through the current UI” (like you noticed).
You can actually get them compared in the UI but it requires specifying things manually on the url. If you look at the URL in the filmstrip view, there is a list of comma separated tests. Each one of those tests can have some options specified that will let you specify which exact run you want to use, a specific label and if you want the first view or repeat view.
-r:X will let you specify a run (where X is the run)
-l:Label will let you specify a label to use
-c:X will let you specify first (X=0) or repeat (X=1) view
The only thing missing is figuring out how to present it to the user for them to select without confusing them (UI was never my strong suit).
I have a suggestion: detect the geo location (using IP address) of the visitor on webpagetest.org and use this to determine the selected/default test location.
When going through the test history, I often see a test run from Dulles first and then from Amsterdam, for Dutch websites.
Obviously, the person running the tests realizes he had the wrong location after viewing the Dulles results.
I’ll take a look and see if I can do something for the default initial setting (once you run a test it remembers your last location setting). The trick will be to do it in a generic way that makes sense for the private installs and can deal with a changing list of locations. Maybe something where I assign each location a set of country codes.
Of course, one reason people may come is to test from a location other than where they are so I may do some research across the tests, requesting IP’s and the URL country before deciding to do anything.
This is incredible!! I’m on my knees! Once again - thank you very much.
Until now I was writing scripts to automate httpwatch to compare the first and the next visit. You have made my life much, much easier.
Regarding thread’s topic:
besides scheduling tests the great enhancement would be support for other browsers. Firefox and chrome were already mentioned here, but as Opera claims to be the most fast of all, comparing Opera with others on the real websites would be very nice.
a small enhancement for those who run private installations would be a support for other than 80 port. At least it is partly supported right now, as test machine can download the job from any http port, but unfortunately it can’t post results to the other than default http port.
And about the scheduled tests: most of the people who require such option are related somehow with IT world. Probably some of them come from technical teams, so they can manage to install and configure the private installations. Maybe this is possible to introduce on hosted version scheduling only once an hour per domain/url, and those who require more frequent tests will be able to run them on privately hosted versions?
I’ve been toying with the idea of allowing registered users to schedule a small number of urls for trended testing - I just haven’t had time to completely think it through yet (so much to do, so little time).
For the port 80 problem, it’s a bug I’ll try to get it fixed in the next release since it has bitten quite a few people.
This probably fits under the heading of “custom headers”, but one thing which would be fantastically useful to those of us using anything like GeoDNS would be if we could specify an IP address as well as a hostname, bypassing the location-specific DNS result and going straight to the specified server regardless. This would be really useful in confirming which geographical DNS rules actually make sense for an existing setup. I guess it’s probably a niche feature… but any chance of a non-UI solution?[hr]
The original intent was for testing dev or staging sites but it can also be used for what you are looking for.
To use a script you need to put an url in the regulalr field (though it won’t be used), select your normal options and then paste a script into the script tab in the advanced section. The commands are tab-delimited so you’ll probably want to edit the script in something like notepad and then paste it in.
That’s prety much the top item on the list for the roadmap We’re planning on working on them in this order:
Chrome
Firefox
Android
The Android support might be worked on in parallel with Firefox because getting a mobile browser into the testing mix is pretty high priority as well. We’re thinking we should be able to have at least basic testing support for most of them by summer.
i would really love to see the following for the test agent machine:
Somehow run the progams (urlBlast, wptdriver, etc.) as services so the agent system can be locked. I am setting up WPT internally so we can test applications that are still in Dev/Test and running an unlocked WPT agent with an admin account, connected to the corporate network without a screensaver is a big NO/NO for our security team …
also, support for Windows 2008 and/or Windows 7 / 64 bit … i’m sure that has already been requested many times
Would they be happier if the machines were actually VM’s inside of a locked ESXi container? The test agents work really well on VMWare so that is an option. I may be able to code around the admin permissions but the unlocked desktop is a necessity for doing actual rendering. We might be able to do something with a headless webkit browser but IE, Chrome and Firefox will need a real desktop for getting useful results (and video capture).
Yep, the only holdup is Dummynet. If you don’t want to do traffic shaping you can run on 64-bit (you could use an external box or VM with a fixed profile if you don’t test different speeds). I have it on my list to get the dummynet driver working on 64-bit windows.
How about the Critical Path feature, that Pagespeed once had as a Beta, and somehow has vanished again?
Additionally I liked the information of from where the object fetch was invoked (i.e. HTML-Tag / example.js / example.css)
If you run a test with Chrome you will get the information about where it was invoked from in the request details (click on an item in the waterfall and look for “Loaded By”.
(like you noticed).
I’m on my knees! Once again - thank you very much.