In discrete event simulations activities may have specified timing properties, such as start time, end time and duration. SubScript itself does not support such notions, but it allows you to plug in a simulation engine that takes care of timing aspects. Suppose there is a variable named sim that represents such an engine; it offers features such as
- currentTime – answers the current time
- startTime – sets the start time of the actions “in here”
- duration – sets the duration of the actions “in here”
Let’s use Daniel Wellman’s library that helps writing down durations, such as in
5.seconds 5 seconds 3 days + 2 weeks
Using a directive we can make sim the simulation engine for the script named live:
main (args: Array[String]) = @sim: live
The directive code sim expands to
sim.apply(here)
Many discrete event simulations focus on queueing behavior. A typical queue simulation is would be:
During a number of days customers enter a shop and queue for service by a server. There are several servers, but there is only one queue with a first-in-first-out rule. The customers arrive according to a random Poisson distribution, and the servers have a random negative-exponentially distributed service time, i.e. number of customers served in unit time. After closing time no more customers are allowed to enter the shop. The shop closes its door after the final customer has been served and left.
A main outline for such a queue simulation with 10 servers and which will simulate a working week, is:
live = days,5 || servers,10 servers (n: Int) = times,n & server,pass server (i: Int) = serveCustomer,i... times (n: Int) = while(pass<n)
The script days takes care of the arrival of customers in time. The or-parallel construct with days,5 guarantees that after 5 simulated days the 10 servers terminate as well, so that the script live halts. The script times is a utility to iterate a given number of times. It activates while(pass<n). The field here.pass expresses the loop counter. pass (without here) refers to a convenience function in the objectsubscript.Predef that returns here.pass; it has pass as an implicit parameter. The script days is simply an iteration of day; day states there are two parallel aspects: the working day and the end of the day, at which a report is generated. During the working day customers are created from 9 to 5:
days (n: Int) = times,n day,pass
day (i: Int) = workingHours,9,17 && atHour,24; theDayEnds,i
workingHours(f: Int,
t: Int) = atHour,f createCustomers
|| atHour,t theShopCloses
workingHours uses or-parallelism to stop the customer creation at closing time. createCustomers creates the customers. This cannot be done by a parallel for loop since the customers should be generated randomly, according to a Poisson distribution, and it is not known how many customers will arrive. We build a sequential iteration that each pass waits for a negative-exponentially distributed time and then launches the customer:
createCustomers = waitTime(randomNegExp 3 minutes) *customer,pass ...
The asterisk in *customer,pass launches the customer script as a separate process, much like the postfix ampersand “&” on the Unix command line.
A customer enters, gets served and leaves. The time he spends in the shop is computed by passing the entrance time et to the script customerLeaves:
customer(c: Int) = var et = sim.currentTime
customerEnters,c
getServed,c
customerLeaves,c,et
The customer getting served and a server providing the service are shared actions, so we define a communicating script:
serveCustomer(s: Int) ,
getServed (c: Int) = var d = (randomNegExp 30 minutes)
@sim.duration = d:
{traceInterval(sim currentTime, sim currentTime+d,
" server "+s+"serves customer "+c)}
Internally this communication first computes the duration, and keeps it in the variable named d. Then a directive tells the simulation engine to apply that duration at the next action. The directive code expands to
sim.duration.apply(here,d)
The other scripts in the simulation are straightforward:
waitTime (d: Duration) = @sim duration = d: η
atHour (h: Int) = @sim startTime = sim currentTime startOfDay + h hours: η
theShopCloses = {! traceTime("the shop closes"); report() !}
theDayEnds (i: Int) = {! traceTime("day "+i+" ends" ); report() !}
customerEnters(i: Int) = {! traceTime("customer "+i +" enters"); nCustomers++ !}
customerLeaves(i: Int,
st: TimeStamp) = {! traceTime("customer " +i+" leaves");
qtime+=sim.currentTime-st !}
The customers in this generation are modelled solely using scripts, not by ordinary objects. Of course you can as well add a customer class and dynamically launch instances; this would make sense in case more statistical data is to be collected:
createCustomers = waitTime(randomNegExp 3 minutes) *new Customer(pass).live...