Note: My Web pages are best viewed with style sheets enabled.

Too Large a Download

Copyright © 2000, 2003-2005, 2010, 2011 by David E. Ross

This was originally written about the excessive sizes of Netscape Communicator versions. (If you liked Netscape, consider upgrading to SeaMonkey, which uses the open-source basis upon which Netscape was developed.) However, this page applies to other software products.

After my employer distributed Netscape 4.7 for use at work late in 1999, I found it to be satisfactory. So I thought I would install it on my home PC. The installation file was over 18 MB. With a 33.6 KBps modem (which is what I had then), that would be over 70 minutes to download — if my modem, Netscape's server, and all intermediate points were to operate at full speed. Instead, because I find that a server connection established via Netscape often fails when I have a long-running download, I downloaded at work via a T1 line. (For long-running downloads at home, I prefer to use WS_FTP.) It still took over 10 minutes, indicating Internet delays that could have increased my download time at home to two hours or more. I then moved the file to a ZIP disc (instead of 13 floppies) to take home.

18 MB is far too large for a download installation file. This one contained too much unnecessary software. Besides basic Netscape components, it included RealPlayer and AIM, products from other companies.

I do want RealPlayer, but I did not need it in Netscape's install file. I already had the latest version installed, downloaded directly from RealNetworks. (I am not even sure Netscape provided the latest version of RealPlayer.) Fortunately, I selected Netscape's "custom install" and was able to suppress the installation of RealPlayer (along with two smaller items). However, that was after it was already downloaded as part of Netscape's installation file. (The latest freeware version of RealPlayer is 26 MB.)

I do not want AIM (AOL Instant Messenger). I refuse to use it. E-mail has the advantage that the recipient deals with it on his own time, not on the sender's time. If I want real-time, live communication over the Internet, I use Skype or the Chatzilla component of SeaMonkey after making and confirming an appointment (via E-mail) for an IRC "meeting". Netscape (owned by AOL, now part of Time-Warner) did not provide a "custom install" option to suppress AIM. I removed it manually (both the files and the registry entries) after installing Communicator.

For several reasons, Communicator should have been broken into smaller modules, each for a different capability (e.g.: Navigator, Messenger, news reader, Composer):

Under generally accepted standards of good software design, modularization is required.
Smaller packages can be downloaded with less risk caused by a failed Internet connection.
Upgrades and patches to a module could be made available without regard to the development schedules of the other modules.
Giving the user a choice of components to download for the "normal install" option eliminates the risk of wrong choices under the "custom install" option.

The Mozilla organization is moving this direction. In addition to the combined SeaMonkey suite (10.6 MB) containing both browser and mail applications, it now has the separate Firefox browser (8.6 MB) and Thunderbird mail application (9.5 MB). Yes, the installation files for Firefox and Thunderbird combined are larger than the installation file for SeaMonkey; but they can be downloaded and installed separately.

On the other hand Adobe Reader is a monolithic 35.0 MB, Skype is 19.8 MB, and RealPlayer is 26.0 MB. Depending on Internet congestion and server loading, these may be slow to download even through a broadband connection. Each of these contains multiple capabilities that should lend themselves to modularization. Then each module could be capable of being separately downloaded and installed.

This should be a warning to all software developers who distribute their products over the Internet.

If your product requires a compressed installation file larger than 10 MB, take another look at its structure. If it is a single monolithic component, it is poorly designed. If it is a set of distinct components that interface with each other, consider distributing it component by component (after freezing the interface) as the Mozilla organization is now doing with Firefox and Thunderbird.
Do not bundle your product with the products of other developers unless your product will not work at all without those other products. Rather than bundling, put links on your distribution Web page to the distribution Web sites where users can obtain those other products. This assures users that they are obtaining the latest, truly authentic versions of those products. Mozilla uses this approach.
The problem does not go away if the installation file tries to download even more files when I go to install your product. Instead, these smaller installer files — stubs — merely delay the problem until installation, which might be several days after I first download what I thought was a complete package.
Making things worse, rarely do these stubs come with a warning in advance that they require an Internet connection to obtain the larger files. Usually, I do my installations and upgrades when I am not connected. That way, as I log each installation, the log is not contaminated by unrelated Intenet activities.
In general, do not deliver your software products in the form of a stub that needs an Internet connection to download larger files. When I install software on my PC, I also want to install the same software on my wife's PC. I want to download the installation only once, not twice.
If the use of a stub cannot be avoided, however, you must place a prominent warning on the download Web page that installation is impossible without an active connection to the Internet.
Consider giving users the option to download via anonymous FTP. While newer browsers have quite robust download capabilities, there are still pathological Web pages that can crash a browser, causing the loss of a partial download. FTP applications, however, operate independently of browsers and are not affected when a browser crashes.
Also, for larger downloads, consider providing a method to check the integrity of the downloaded file. Symatec provides MD5 checksums for its virus definition update files. The Mozilla Foundation provides several checksums for its product installation files, including MD5. (While the MD5 algorithm is generally no longer used for encryption because it was found to contain some vulnerabilities against hostile attacks, it is still useful for checking large files against accidental corruption.)

Remember, about 10%-20% of those who access the Internet from home still use dial-up modems. Even through a 56 Kbps modem (the fastest available), 20 MB still takes over 45 minutes to download — if your server can indeed deliver 56 Kbps. If your server cannot deliver 56 Kbps, a broadband connection (e.g., DSL, cable modem, T1) would not help me; in this case, you should not ask anyone to download megabyte files from your site.

5 February 2000
Updated 3 January 2011