This post takes a closer look at Base64 image performance, offers some use cases and raises some questions. Iâ€™ve had a number of conversations recently over the benefits of using client-side, Base64 image strings to improve web app performance over making multiple <img> requests. So I ran a few tests of my own, and the results are shown at the bottom of the post.
There are other ways of doing this that Iâ€™m not covering in this post such as using server-side code to convert images on the fly, passing Base64 strings in URLs, etc.
Size? Base64 image strings will always be larger than their native counterparts. Take the following example using a copy of the relatively small HTML5 logo and a decent sized screenshot. As you can see the text equivalent is 33% and 31% greater, respectively.
Html5.png Â = 1.27KB (64×64 pixels)
Html5base64.txt = 1.69KB (1738 characters)
% Difference =Â +33%
Screenshot.png = 19KB (503×386 pixels 24bit)
ScreenshotBase64.txt = 24.9KB (25584 characters)
% Difference = +31%
In comparison, when working with <img> tags you’ll be working with an ID that points to the actual image file stored in memory.
Convenience? As you can see, the length of your base64 strings can get quite long. The simple HTML5 logo in the previous example becomes a 1738 character long string and thatâ€™s only 1.69KBs worth of image.
Can you imagine having a dozen of images similar in size to the 19KB Screenshot example? That would create over 300,000 ASCII characters. Letâ€™s put that into the perspective of a Word document. Using 1â€ť margins all the way around, this would create a document approximately seven and a quarter pages long!
I assert that Base64 is best for static images, ones that donâ€™t change much at all over time. The bigger the image the more time consuming it can become convert it, copy-and-paste it into your code and then test it. Any time you make a change to the image youâ€™ll have to repeat the same steps. If you accidentally inject a typo into a Base64 string you have to reconvert the image. Thatâ€™s simply the best approach from a productivity perspective.
In comparison when using a regular old PNG file, you create the new version, copy it out on the web server, flush your browser cache, run a quick test with no need to change any code and bang youâ€™re ready to go have a cup of coffee.
100,000 page hits/ month (main.html) x 256KB = 25.6 GB (incls. 75KB of standard PNG images)
100,000 page hits/month (main.html) x 293.5KB = Â 29.4 GB (incls. 97.5KB Base64 images)
Also, some providers give you decent tools that you can use to experiment with Base64 images versus regular images and test that against your bandwidth consumption.
Latency? This variable doesnâ€™t really apply directly to Base64 images. There are many factors that determine latency that I’m not going to discuss here. There are some more advanced networking tools that let you figure out average latency on your own web servers. Every request will be unique based on network speed, the number of hops between the client and the web server, how the HTTP request was routed over the internet, TCP traffic over the various hops, load on the web server, etc.
A few quick performance tests.
These tests were performed on a Chrome Browser over a CenturyLink DSL with a download speed of 9.23MB/sec and an upload speed of 0.68MB/sec. SeveralÂ tracert(s)Â of TCP requests from my machine to the web server showed more than 30 hops with no significant delays or reroutes. The web server is a hosted machine.
Time to load 1.9KB HTML file: 455ms
Total load time: 1.64s
Time to load cached HTML file: 293ms
Total load time: 460ms
Test 3 – using <img> tags to request PNG images. This test consists of an uncached HTML file that contains five <img> tags pointing to five remotely hosted 20KB PNG files.
Time to load 1KB HTML file: 304ms:
Time to load five images: 776 ms
Total load time: 1.08s
Test 4 – cached html file using <img> tags to request PNG images. This consists of reloading Test 3 in the browser
Time to load cached HTML file: 281ms
Time to load five cached images: 16ms
Total load time: 297ms
It’s not 100.0000% true that multiple HTTP requests results in slower application performance in comparison to embedding Base64 images. In fact, I’ve seen anecdotal evidence of this before on production apps, so it was fun to do a some quick testing even if my tests were not completely conclusive beyond a doubt.
My goal was to spark conversation and brainstorm on ideas. I know some of you will say that thousands of tests need to be run an statistically analyzed. My argument is that these tests represented actual results that I could see with my own eyes rather than being lumped into some average or medium statistic.
Note that I’m just posting a snapshot of the tests I ran. I didn’t have enough time to draw up a significant battery of tests to cover as many contingencies as possible. However, the test numbers I’ve posted were fairly consistent in the favor of the multiple PNG requestsÂ loadingÂ faster than a single .js file containing five Base64 images. Obviously more significant testing is needed to sort out other real-world variables, such as image file sizes versus application size and under a variety of conditions and different browsers.
[Edited 2/26/13: fixed a few typos]