Folks,

How to Split a Server, part 3

I've posted part 3 of my article series, How to Split a Server on my website, and it should be up on http://afp548.com/ shortly. It covers separating out the e-mail and web parts of the example.org domain from the example.com domain, including forwarding old e-mail addresses, having the same address independently for both domains, and how to organize your web server's file system.

http://ps-enable.com/articles/how_to_split_a_server_part_3.html

How NOT to Test a System

The Diebold electronic poll book system is designed to check in registered voters at a precinct -- not actually to record votes, but to make sure that a voter gets to cast only one ballot. It had a number of serious problems in the September 12 primary, but the egregious one that makes me really steamed is the one that Diebold supposedly just fixed. From the Washington Post article: 

http://www.washingtonpost.com/wp-dyn/content/article/2006/09/25/AR2006092501480.html

"Diebold technicians showed what happened to an unmodified e-poll book after 40 to 50 voters had registered: An error message appeared on the screen, displaying the words "can not continue;" the screen went black; and the unit rebooted, as if a polling judge had just turned it on."

OK, this tells me that the software was never stress-tested -- this was a consistent, repeatable bug, the kind that is relatively easy to find and to fix. I can't believe that this was not caught -- 40-50 voters is not all that many at a given precinct. Since there are 5500 electronic poll book units and turnout for the last primary (2004) was 646,000 people, this gives a an average of 117 voter check-ins per electronic poll book unit. (That's an optimistically low number, by the way -- the voter turnout in the 2002 primary was 2.76 million.) The fact that the poll book units failed after less than half of that number is a sign that the units were not stress tested.

"Then, Diebold technician Euel Kirk Cowal began registering voters on a unit with the modified software. Cowal reached 138 voters without a freeze. The Diebold executives, including one from the company's Texas headquarters and another from a subsidiary in California, looked relieved."

OK, so at least one test unit exceeded the likely primary numbers. We would want to repeat the test multiple times to make sure that it's really fixed, but we'll give them that much. What about the general election? If you run the numbers, each electronic poll book should be tested to handle around 2500 voters check-ins. This is not that many voter check-ins by any standard.

In other words, the software has been tested to 1/18th of the level it should be tested to. Bleah. That system should never have been deployed for the current election -- it hasn't passed reasonable acceptance tests.

Request for Feedback

This is now the tenth newsletter that I've sent out, and I'd like to get some feedback from you. What articles did you like or not like? Are there topics that you would like me to cover that I haven't so far? If you would like to see some of the back issues, all of my newsletters are archived on my website at http://ps-enable.com/articles/newsletters/.


--Paul


Paul Suh                                                         http://www.ps-enable.com/
paul.suh@ps-enable.com                           (240) 672-4212