Reader Randy King recently performed an unusual experiment that provided some really good end-of-the-year food for thought:
I have an old Gateway here (120 MHz, 32 Mbytes RAM) that I "beefed up" to 128 Mbytes and loaded with--get ready--Win 95 OSR2. OMIGOD! This thing screams. I was in tears laughing at how darn fast that old operating system is. When you really look at it, there's not a whole lot missing from later operating systems that you can't add through some free or low-cost tools (such as an Advanced Launcher toolbar). Of course, Win95 is years before all the slop and bloat was added. I am saddened that more engineering for good solutions isn't performed in Redmond. Instead, it seems to be "code fast, make it work, hardware will catch up with anything we do" mentality.
It was interesting to read about Randy's experiment, but it started an itch somewhere in the back of my mind. Something about it nagged at me, and I concluded there might be more to this than meets the eye. So, in search of an answer, I went digging in the closet where I store old software.
Factors Of 100
It took some rummaging, but there in a dusty 5.25" floppy tray was my set of install floppies for the first truly successful version of Windows--Windows 3.0--from more than a decade ago.
When Windows 3.0 shipped, systems typically operated at around 25 MHz or so. Consider that today's top-of-the-line systems run at about 2 GHz. That's two orders of magnitude--100 times--faster.
But today's software doesn't feel 100 times faster. Some things are faster than I remember in Windows 3.0, yes, but little (if anything) in the routine operations seems to echo the speed gains of the underlying hardware. Why?
The answer--on the surface, no surprise--is in the size and complexity of the software. The complete Windows 3.0 operating system was a little less than 5 Mbytes total; it fit on four 1.2-Mbyte floppies. Compare that to current software. Today's Windows XP Professional comes on a setup CD filled with roughly 100 times as much code, a little less than 500 Mbytes total.
That's an amazing symmetry. Today, we have a new operating system with roughly 100 times as much code as a decade ago, running on systems roughly 100 times as fast as a decade ago.
By itself, those "factors of 100" are worthy of note, but they beg the question: Are we 100 times more productive than a decade ago? Are our systems 100 times more stable? Are we 100 times better off?
While I believe that today's software is indeed better than that of a decade ago, I can't see how it's anywhere near 100 times better. Mostly, that two-orders-of-magnitude increase in code quantity is not matched by anything close to an equal increase in code quality. And software growth without obvious benefit is the very definition of "code bloat."
Instead, most of today's software is produced with high-level programming languages that often include code-automation tools, debugging routines, the ability to support projects of arbitrary scale, and so on. These tools can add an astonishing amount of baggage to the final code.
This real-life example from the Association for Computing Machinery clearly shows the effects of bloat: A simple "Hello, World" program written in assembly comprises just 408 bytes. But the same "Hello, World" program written in Visual C++ takes fully 10,369 bytes--that's 25 times as much code! (For many more examples, see http://www.latech.edu/~acm/HelloWorld.shtml. Or, for a more humorous but less-accurate look at the same phenomenon, see http://www.infiltec.com/j-h-wrld.htm. And, if you want to dive into Assembly language programming in any depth, you'll find this list of links helpful.)
Human skill also affects bloat. Programming is wonderfully open-ended, with a multitude of ways to accomplish any given task. All the programming solutions may work, but some are far more efficient than others. A true master programmer may be able to accomplish in a couple lines of Zen-pure code what a less-skillful programmer might take dozens of lines to do. But true master programmers are also few and far between. The result is that code libraries get loaded with routines that work, but are less than optimal. The software produced with these libraries then institutionalizes and propagates these inefficiencies.
You And I Are To Blame, Too!
All the above reasons matter, but I suspect that "featuritis"--the tendency to add feature after feature with each new software release--probably has more to do with code bloat than any other single factor. And it's hard to pin the blame for this entirely on the software vendors.
Take Windows. That lean 5-Mbyte version of Windows 3.0 was small, all right, but it couldn't even play a CD without add-on third-party software. Today's Windows can play data and music CDs, and even burn new ones. Windows 3.0 could only make primitive noises (bleeps and bloops) through the system speaker; today's Windows handles all manner of audio and video with relative ease. Early Windows had no built-in networking support; today's version natively supports a wide range of networking types and protocols. These--and many more built-in tools and capabilities we've come to expect--all help bulk up the operating system.
What's more, as each version of Windows gained new features, we insisted that it also retain compatibility with most of the hardware and software that had gone before. This never-ending aggregation of new code atop old eventually resulted in Windows 98, by far the most generally compatible operating system ever--able to run a huge range of software on a vast array of hardware. But what Windows 98 delivered in utility and compatibility came at the expense of simplicity, efficiency, and stability.
It's not just Windows. No operating system is immune to this kind of featuritis. Take Linux, for example. Although Linux can do more with less hardware than can Windows, a full-blown, general-purpose Linux workstation installation (complete with graphical interface and an array of the same kinds of tools and features that we've come to expect on our desktops) is hardly what you'd call "svelte." The current mainstream Red Hat 7.2 distribution, for example, calls for 64 Mbytes of RAM and 1.5-2 Gbytes of disk space, which also happens to be the rock-bottom minimum requirement for Windows XP. Other Linux distributions ship on as many as seven CDs. That's right: Seven! If that's not rampant featuritis, I don't know what is.
But there are signs that we may have reached some kind of plateau with the simpler forms of code bloat. For example, with Windows XP, Microsoft has abandoned portions of its legacy support. With fewer variables to contend with, the result is a more stable, reliable operating system. And over time, with fewer and fewer legacy products to support, there's at least the potential for Windows bloat to slow or even stop.
Linux tends to be self-correcting. If code-bloat becomes an issue within the Linux community, someone will develop some kind of a "skinny penguin" distribution that will pare away the needless code. (Indeed, there already are special-purpose Linux distributions that fit on just a floppy or two.)
While it's way too soon to declare that we've seen the end of code bloat, I believe the signs are hopeful. Maybe, just maybe, the "code fast, make it work, hardware will catch up" mentality will die out, and our hardware can finally get ahead of the curve. Maybe, just maybe, software inefficiency won't consume the next couple orders of magnitude of hardware horsepower.
What's your take? What's the worst example of bloat you know of? Are any companies producing lean, tight code anymore? Do you think code bloat is the result of the forces Fred outlines, or it more a matter of institutional sloppiness on the part of Microsoft and other software vendors? Do you think code bloat will reach a plateau, or will it continue indefinitely? Join in the discussion!