User talk:SMcCandlish/sandbox quoted ref test
Conclusion
[edit]Despite repeatedly giving benefit of the doubt, if not outright favoritism, toward "lazy" formatting rather than proper XML syntax, page loading was still most often noticeably lower without the quotation marks, a result that can only be attributed to the additional MediaWiki parser time required to compensate for substandard syntax.
Because of the intentional benefit-of-the-doubt bias toward the unquoted markup, the estimate of how much server overhead is blown on compensating for that markup is a very conservative one.
Possible further testing
[edit]If a bot were to run these tests every hour on the hour for several months, and simply average it all out, we would expect to see an even higher percentage difference in favor of quoted markup. If the bot also discarded test pairs with excessive differences, the manual results I encountered indicate that better efficiency of the quoted, proper-XML markup would likely be proven independently by three different statistical factors in the bot's data (in decreasing order of statistical relevance and reliability):
- the calculated averages of the pages' load time in the accepted data;
- the extent to which times that favor quoted values do so, vs. the extent to which those that favor unquoted do; and
- the frequency with which the discarded data favored quoted vs. unquoted values.
Other tools can be used, but care must be taken to distinguish those that do not include rendering time, only delivery time, such as Tools.Pingdom.com.
— SMcCandlish ☺ ☏ ¢ ≽ʌⱷ҅ᴥⱷʌ≼ 20:18, 27 June 2015 (UTC)