On 4/5/07, Jonathan Vanasco <[EMAIL PROTECTED]> wrote:
Right. My point is that no matter what, you're going to have multiple children locking and blocking against each other. The efficiency of the operations are negligible when compared to one another , and the existence of blocking makes it virtually insignificant.
Although that sounds reasonable, when I've tried to test it my results have been the opposite -- storage speed really matters and locking doesn't matter much at all. I did some benchmarking of various cache modules for a talk I gave at OSCON in 2002. At the time, I found that the number of HTTP processes trying to write to the cache didn't change the overall performance picture, i.e. the ones that used the fastest storage mechanisms clearly won both in read-only and write-heavy situtations. It might be because the locks are held for shorter periods of time, or that the number of mod_perl processes competing for locks is not all that high, or possibly just a lame test, but I couldn't get the locking to register as significant. Massive grain of salt not included. - Perrin
