Jakob Nielsen is positive that we should achieve sub-8 seconds load time for the users to enjoy navigating our site. Another reason for speed optimisation is reducing the load for sites with heavy traffic. In the above report, it is obviously a rather impossible task to scale down 120 kilobyte (20 kilobyte of CSS are omitted by the analysis) to 30, necessary for sub-8 load time, but we might want to get closer to that usability ideal. (Side note: at that moment bitweaver was using Ajax for some packages - 45k!)
eliminate extra HTTP requests
Reading: Modem connections (56Kbps or less) are corrected by a packet loss factor of 0.7. All download times include delays due to round-trip latency with an average of 0.2 seconds per object. With 31 total objects for this page, that computes to a total lag time due to latency of 6.2 seconds. In the above example, we are loosing more then 6 seconds just for HTTP requests, so there's no chance for sub8 response :(
"The less requests the better", so before optimizing file size, think how to reduce quantity of files per page.
Think twice before you start "slicing" into small pieces your images and backgrounds in some Adobe ImageReady. This technique is an offspring of table-based layouts — when bw has div-based layout at the moment. IMHO with div-based layout you might get better results having 3 big background images instead of 30 small ones.
Also, some of the images can be eliminated without any damage to the style. There's a nice CSS trick for rollover images (if using graphic rollover at all): the rollover images are stacked in single background file vertically, and on :hover the vertical position of background is changed in CSS. The same can be done with very small images, like icons and signs of external link etc. The trick is that it doesn't give any boost in load time if the file is smaller then 1160 bytes - it's still single TCP-IP packet. So we can combine two small graphic files if their total is still less in size then 1k. This way we cut off 1 packet and 1 HTTP request. (At this point, reading about how much effort is needed to save 0.2 seconds, all readers are confident I'm crazy and need psychiatric attention %/ )
Combine javascript and CSS files: put all external CSS into one file - less files means less HTTP requests.
Eliminate redundant markup before uploading to the server, but leave read-frendly version locally for editing.
170k images! (which also means the actual frame is rather slick)
WebWait (11/2007)
webwait.com: Use WebWait to benchmark your website or test the speed of your web connection. Timing is accurate because WebWait pulls down the entire website into your browser, so it takes into account Ajax/Javascript processing and image loading which other tools ignore.
Link to page analyzed
Average load time after 5 runs (in seconds, call interval: 60 seconds)
with some further qualitative analysis : there are green points such as use of Gzip, only one HTML and one CSS file, but too many images (18), and too much Javascript (71k). Of course, nobody today still uses 56k links, but still at T1 speed the calculated speed is about 6 seconds (Yes, I too get more than 9 by simply adding, go figure). So a remommendation would be to use Gzip also for javascript packages. Also adding width and height attributes to all images should make rendering faster.
What could be done?
eliminate extra size
Jakob Nielsen is positive that we should achieve sub-8 seconds load time for the users to enjoy navigating our site. Another reason for speed optimisation is reducing the load for sites with heavy traffic. In the above report, it is obviously a rather impossible task to scale down 120 kilobyte (20 kilobyte of CSS are omitted by the analysis) to 30, necessary for sub-8 load time, but we might want to get closer to that usability ideal.
eliminate extra HTTP requests
Reading: Modem connections (56Kbps or less) are corrected by a packet loss factor of 0.7. All download times include delays due to round-trip latency with an average of 0.2 seconds per object. With 31 total objects for this page, that computes to a total lag time due to latency of 6.2 seconds. In the above example, we are loosing more then 6 seconds just for HTTP requests, so there's no chance for sub8 response :(
"The less requests the better", so before optimizing file size, think how to reduce quantity of files per page.
Think twice before you start "slicing" into small pieces your images and backgrounds in some Adobe ImageReady. This technique is an offspring of table-based layouts — when bw has div-based layout at the moment. IMHO with div-based layout you might get better results having 3 big background images instead of 30 small ones.
Also, some of the images can be eliminated without any damage to the style. There's a nice CSS trick for rollover images (if using graphic rollover at all): the rollover images are stacked in single background file vertically, and on :hover the vertical position of background is changed in CSS. The same can be done with very small images, like icons and signs of external link etc. The trick is that it doesn't give any boost in load time if the file is smaller then 1160 bytes - it's still single TCP-IP packet. So we can combine two small graphic files if their total is still less in size then 1k. This way we cut off 1 packet and 1 HTTP request. (At this point, reading about how much effort is needed to save 0.2 seconds, all readers are confident I'm crazy and need psychiatric attention %/ )
Combine javascript and CSS files: put all external CSS into one file - less files means less HTTP requests.
Eliminate redundant markup before uploading to the server, but leave read-frendly version locally for editing.
170k images! (which also means the actual frame is rather slick)
WebWait (11/2007)
webwait.com: Use WebWait to benchmark your website or test the speed of your web connection. Timing is accurate because WebWait pulls down the entire website into your browser, so it takes into account Ajax/Javascript processing and image loading which other tools ignore.
Link to page analyzed
Average load time after 5 runs (in seconds, call interval: 60 seconds)