<< Use a coordinated transactor | Inserting a Synchronization Point >>

A Negative Exponential distribution

Inserting Other Items into a Suite

285

A Uniform distribution means that over time, the transactions average out to the
rate you specify, although the time between each transaction is constant. The time
between the start of each transaction is chosen randomly with a uniform
distribution within the selected range. Think of this range as a "window" through
which the transaction runs.

For example, the transaction rate is 4 per minute (that is, 1 transaction per
15-second interval). If you select a range of 20%, your transaction has a 3-second
window on each side of that 15-second interval, because 20% of 15 seconds is 3
seconds.

Therefore, the first transaction starts at 1218 seconds (15 plus or minus 3). The
second transaction starts 15 seconds plus or minus 3 seconds after the first
transaction starts. If the first transaction starts at 12 seconds, the second transaction
starts at 24 to 30 seconds. However, if the first transaction starts at 18 seconds, the
second transaction starts at 30 to 36 seconds.

Because each transaction starts randomly within the range that you specify, it is
normal for transactions to run at a rate that is faster or slower than the rate that
you selected for short periods of time. For example, if a transaction starts every 12
seconds for a minute (recall that the window is 1218 seconds), the rate for that
initial interval is 5 per minute--not the 4 per minute that you selected. Over time,
however, the transaction rate averages out to 4 per minute.

With a Uniform distribution, a transaction has the same probability of running
within the range that you specify. The transaction starts anywhere within this
window. In our example, the probability of the first transaction starting at 12
seconds, 18 seconds--or anywhere in between--is equal.

A Negative Exponential distribution, in contrast, changes the probability of when a
transaction starts. This distribution most closely emulates the bursts of activity
followed by a tapering off of activity that is typical of user behavior. Using the
same example of 4 transactions per minute, the probability that a transaction starts
immediately is high, but decreases over time. TestManager maintains the desired
average rate.

Imagine that you have called a meeting at two o'clock. Most people arrive at two, a
few people arrive at five minutes past two, and fewer still at ten past two. Perhaps
the last straggler arrives at two-thirty. This arrival time approximates a negative
exponential distribution. Most people arrive on time, and then the arrival rate
declines. Mathematically speaking, the interval is chosen randomly from a
negative exponential distribution with the average interval is equal to 1/rate.