Allan Wind: > On 2005-05-29T11:51:27-0400, John Ruttenberg wrote: > Content-Description: message body text > > I have a benchmark program that demonstrates significant performace > > degradation over time. Here is what this benchmark does: > > > > 1. Create a table with a primary integer key and a blob value. > > 2. Populate the table with 1M rows. In all cases the blobs are strings > > with random lengths between 1 and 1k bytes. > > Does your random generator run in constant time? >
I timed it. It takes just under a second for 1M calls to rand(). This didn't change over thousands of repetitions. > > 3. Loop foreer, making passes. A pass does 1M operations each of which > > is randomly selected and can be either: > > a. an insert of an existing row, chosen at random, or > > You mean an update? How do you select a random row? In particular is > that a constant (in time) operation? > No, sorry. I meant a *delete* of an existing row. The choice takes almost no time. > > b. an assert of a new row as in 2. > > Not sure what that means. If you select a random number, then test for > pre-existence of the row, then the chance of conflict with go up with > time. > Again, sorry. I posted too fast. I meant an insert of a new row. > > Here are plots of time for each pass on a modern linux machine (fc3, >3GHz > > processor, ide drives), and a modern XP machine (similar hardware). > > You need to label the axis and include dimension. > > The plots show iteration number on the x axis and number of secons on the y axis. So the first iteration takes about 50 seconds on linux and by the 300th iteration, it's taking more than 600 seconds.
