Is Your Site Too Slow? Why Image Optimization Is Critical

This post was originally published on May 13th, 2004. As others are writing about the topic, I thought bringing it out of the archives would be worthwhile.

A little recap

The idea of placing multiple states of buttons and other elements that are used in background images took its roots, I believe, from Pixy's Fast Rollovers. The CSS Zen Master extended this to another purpose in CSS Sprites: Image Slicing’s Kiss of Death. Didier Hilhorst came up with a nice application of this method, and I worked it backwards in Responsible CSS - Recycle your background images.

The idea behind the 'sprites' method can obviously be extended to any html element, and there are tangible benefits for doing this, just as long as the designer does his or her usual homework.

Benfits of using the 'sprites' method

What are the possible the benefits of using this method? Essentially it lies in faster download times for your web content.

Readers of Andy Kings book, Speed Up Your Site: Web Site Optimization will notice that this method reduces http requests and makes more efficient use of the data packets used to transfer files to the users computer, and that that is a good thing.

Packet size and http requests

From Web Page Design and Download Time, by Jing Zhi of Keynote Systems (seen here - pdf), cited in Andy's book:

The basic performance principle is therefore to make fewer requests and transmit fewer packets. From this principle, we can derive two basic design rules for wellperforming Web pages. First, reduce the overall size of the page, thereby reducing the number of bytes (and packets) to be transferred over the Internet. Second, limit the number of embedded objects on the page, such as images, each of which must be requested and transferred separately from server to browser.

They also found that it was the number of packets and not necessarily the overall size of the page that was important. If a packet could hold 1460 bytes (the figure given in the article) and your object was 1600 bytes, it would require two packets. They found that this object would transfer at the same speed as another object that was greater in size but still fit in two packets.

Potential payoff

The potential payoff for using this method versus individual images, then, is a faster download time due to reduced number of packets and fewer http requests.

Reducing http requests is easy. One file instead of two or three etc. is simple. But packet requests? That depends...

An example

The number of packets sent will depend on the size of the file and the users internet connection.

As an example, lets look at the fiftyfoureleven.com logo at the top of the page. When this design was first being coded, that link consisted of two 3.34kb images, one for the link state and one for the hover state. Now, by using one image that contains both states and simply bumping it back and forth depending on the hover state, that has been reduced to one 5.35 kb image. Right there is a savings of 1.33 kb. Good news.

Now, for arguments sake lets say that a packet can hold 1460 bytes (packet size for connections greater than 128kb/s = 1500 bytes -40bytes for tcp/ip headers). The two image method used 6 packets, 3 for each image (3.34/1.46, rounded up). The single image method uses 4 packets (5.34/1.46, rounded up).

Things are looking good.

How to optimize

In his alistapart article, Dave refers to the image that holds all of the sprites as his 'master image'. The key to benefitting from this method is to ensure that the file size of your master image isn't a bloated equivalent versus the sum of its pieces.

Conclusion

Great benefits can be realized when combining a master image from slices that fall well below the size of one packet, as that unused packet space goes wasted.

After doing a little more research, it seems that packet size can vary depending on the connection rate. That being said, it may be rather difficult to come up with a firm rule here. To play it smart and safe, try and:

• build master images that are smaller then the sum of their collective slices (by combining images of similar colors, for example) or
• rather than use a different distinct image for a certain element, reuse one that you already plan on using elsewhere.

This isn't exactly groundbreaking advice, however having seen the results acheived with the logo on this page, it can be seen that using the sprite method versus individual images at minimum does reduce http requests and even further it can reduce file size which in turn can reduce the number of packets sent.