WebPagetest Forums

Pat!
Hi there! Long time no see! :-) I am Markus Leptien from AOLs former german subsidiary. Came most times along with Volker Hochstein!

Awesome tool! It actually puts two tools together I was currently using. It has the ability to watch objects over SSL and correlate them to TCP Connections. Formerly I was using HTTPWatch for looking at objects via SSL, but is unable to have a connection view.
And Microsofts VRTA, which gives me the connections, but is unable to distinguish the objects via SSL.

So, the http://www.alice-dsl.de queries is actually me.
Visible there and on our roadmap are the following improvements:
-Concatenate the CSS Files to probably just 2
-Reduce JS Files in the Head to 1
-Put all other JS Files at the bottom, to be downloaded and in parallel, after Rendering complete
-Expires Headers 10 yrs in the future for CSS, JS, GIF and JPG, while injecting the Last-modified into the URL.
-Put Omniture and ivw-Counters after Rendering complete
-Reduce image-weight, especially with the 45K Header Image
-Put CSS, JS, GIF and JPG on a new domain static.alice-dsl.de

Anything else you would suggest?

Btw, 2 other Questions:
a) How do you do that? :-) Especially that it works via SSL.
b) Is there Firefox Release on the Roadmap? :-)

Best wishes,
Markus

Hey, great to hear from you - as soon as I saw the LeptienM account I knew exactly who it was :-)

Ouch, I just looked at the test results and it looks pretty painful. The css and js concatenation should be priority number 1 and will radically improve the times (particularly the start render time). After you do that and get some decent expires headers in it's probably worth re-testing to see where the pain points are.

Be careful with breaking out too many domains. With SSL the cost for setting up additional connections is pretty steep so you'll want to make sure it is worth it (and once you combine the files it might not be).

On the "How it's done" side, I have a browser plugin that hooks a bunch of different API's and intercepts the calls (both into WinInet and Winsock). It is pretty similar to how HTTPWatch does things. The connection view was added after talking to the guys who created VRTA :-)

Firefox is on the roadmap but it is probably quite a ways out as it would be a complete rewrite of the browser plugin (not something I am looking forward to doing).

In case you haven't seen it yet, I just added the ability to record a video capture of the test runs if you want to see what it looked like loading. I've attached a sample of the Alice page loading from Dulles. It gets particularly interesting if you want to compare a "before" and "after" or compare to a competitor. Here are more details on it: http://blog.patrickmeenan.com/2009/09/re...etest.html

Great to hear from you!

-Pat

Yep, seems like somebody took the preambel in chapter 1 of Even Faster Websites pretty serious:
Premature optimization is the root of all evil
:-)
So this page went live (actually a whole new CMS was introduced) and people are wondering, why it's so slow. So Optimization hasn't being done at all up until today. But due to your magnificent tool its pretty visible, were the culprit is.

Regarding FF, I saw that your traffic is 94% IE7 and IE8. "Unfortunately" here with us its different. On our page we have ~45% IE7 and IE8, but also 40% FF3.x. Will keep you posted, as optimization progresses. Next staging, which includes first fixes, should be roughly in a month.

Best wishes and also greetings from Volker,
Markus

P.S.: Its funny. Blank pages in a movie leave a much more shocking impression, compared to when I am surfing this site :-)

Hej Pat!

We have added now Expires 24hrs. in the future and a Cache-Control: Max-age 86400 to all objects.
Even though load time for 2nd view improved quite a bit in your test from roughly 4,4 secs. to 2,7 secs. as expected.
Nevertheless the waterfall chart still shows something like 63 requests being done in repeat view, even though the Headers are correctly having the Cache advice. Any idea, why the tests still shows these requests in Repeat view? Test is here:
http://www.webpagetest.org/result/090916_28H7/

Best wishes,
Markus

Looks like a session ID is being appended to the urls which makes them different objects.

the gzip compression is not working correctly on the webserver
https://www.alice-dsl.de/selfcare/conten...8.jvmcae01 returns the uncompressed document while https://www.alice-dsl.de/selfcare/conten...y-1.3.2.js returns the document as compressed (so dropping the sessionid would allow compression + cache)

@pmeenan: webpagetest doesn't detect https://www.alice-dsl.de/selfcare/conten...8.jvmcae01 as uncompressed text response

also mod_concat might help with those man css and js files

Unfortunately, Gzip checking doesn't work for SSL requests so it is disabled (and don't trust the headers you see with regards to gzip for SSL requests).

Long version:

For the SSL requests I don't have visibility at the raw socket level and IE asks WinInet to automatically decompress the requests (and WinInet is kind enough to modify the headers so I can't tell they were gzipped). I can get it working for IE7 by forcing WinInet to not decompress the responses but IE8 breaks because it can't natively deal with gzipped responses

I have a change in dev that enables the gzip checking for https on IE7 only (nothing I can do short-term about IE8). I want to test it a fair bit and also look at a couple of other things while I have the hood open but I'll let you know when it is live.

Thanks,

-Pat

It's only live in Dulles at this point but checking gzip for https is now working with IE7. Here are the results: http://www.webpagetest.org/result/090922_2A3A/

It is correctly picking up that your jquery files are not being gzipped.

Thanks,

-Pat

Hi!

So, we have done most of the stuff, and the results are pretty impressive when tested from Germany. As my company is a german Telco, performance is mainly relevant to german customers.

Time to Fully Downloaded with an empty Cache decreased from 6.8 seconds to 2.8 seconds.
Time to Fully Downloaded with a primed Cache decreased from 3.3 to 0.8 seconds.

Number of HTTP Requests decreased from 89 to 48 with an empty Cache.
Number of HTTP Requests decreased to 4 with a primed Cache (Dynamic Basepage + 2 Counting Pixels + 1 Ad).

Number of Bytes downloaded decreased from 584K to 294K.

What we did was:
Put all static assets on a cookie-free domain.
Concatenate CSS Files
Concatenate JS Files and split the initial load
Sprite images
Have far future Expires
GZIP everything
Have Cache-Control: Public enabled for Browser Caching of SSL content

There is still a couple of stuff we would like to do.
Which is introduce a second static Domain, to enable SSL Session Caching and to unblock loading of the big Javascript file in the beginning.
Maybe a little bit more spriting could be done as well. Also we will introduce some serverside caching of the static assets, which are currently delivered from the CMS each and every time.

Apart of that it is indeed impressive, what you can gain "for free" with a little bit of dev effort/config tweaking. And the situation is a clear win-win: Faster Page, less load on the infrastructure.

Pat, thanks a lot for your help and this tool. I hope that I/we will set up an own server instance of your tool in 2010. And maybe I can convince people here to make the instance public.

Have a great X-mas
Markus