These examples are adapted from the Python package ‘SimPy’, here. Users
familiar with SimPy may find these examples helpful for transitioning to
simmer
. Some very basic material is not covered. Beginners
should first read The Bank Tutorial:
Part I & Part II.
Covers:
From here:
A carwash has a limited number of washing machines and defines a washing process that takes some time. Cars arrive at the carwash at a random time. If one washing machine is available, they start the washing process and wait for it to finish. If not, they wait until they an use one.
Parameters and setup:
library(simmer)
NUM_MACHINES <- 2 # Number of machines in the carwash
WASHTIME <- 5 # Minutes it takes to clean a car
T_INTER <- 7 # Create a car every ~7 minutes
SIM_TIME <- 20 # Simulation time in minutes
# setup
set.seed(42)
env <- simmer()
The implementation with simmer
is as simple as defining
the trajectory each arriving car
will share and follow.
This trajectory comprises seizing a wash machine, spending some time and
releasing it. The process takes WASHTIME
and removes some
random percentage of dirt between 50 and 90%. The log_()
messages are merely illustrative.
car <- trajectory() %>%
log_("arrives at the carwash") %>%
seize("wash", 1) %>%
log_("enters the carwash") %>%
timeout(WASHTIME) %>%
set_attribute("dirt_removed", function() sample(50:99, 1)) %>%
log_(function()
paste0(get_attribute(env, "dirt_removed"), "% of dirt was removed")) %>%
release("wash", 1) %>%
log_("leaves the carwash")
Finally, we setup the resources and feed the trajectory with a couple of generators. The result is shown below:
env %>%
add_resource("wash", NUM_MACHINES) %>%
# feed the trajectory with 4 initial cars
add_generator("car_initial", car, at(rep(0, 4))) %>%
# new cars approx. every T_INTER minutes
add_generator("car", car, function() sample((T_INTER-2):(T_INTER+2), 1)) %>%
# start the simulation
run(SIM_TIME)
#> 0: car_initial0: arrives at the carwash
#> 0: car_initial0: enters the carwash
#> 0: car_initial1: arrives at the carwash
#> 0: car_initial1: enters the carwash
#> 0: car_initial2: arrives at the carwash
#> 0: car_initial3: arrives at the carwash
#> 5: car_initial0: 86% of dirt was removed
#> 5: car_initial1: 50% of dirt was removed
#> 5: car_initial0: leaves the carwash
#> 5: car_initial1: leaves the carwash
#> 5: car0: arrives at the carwash
#> 5: car_initial2: enters the carwash
#> 5: car_initial3: enters the carwash
#> 10: car_initial2: 59% of dirt was removed
#> 10: car_initial3: 85% of dirt was removed
#> 10: car_initial2: leaves the carwash
#> 10: car_initial3: leaves the carwash
#> 10: car0: enters the carwash
#> 10: car1: arrives at the carwash
#> 10: car1: enters the carwash
#> 15: car0: 98% of dirt was removed
#> 15: car1: 96% of dirt was removed
#> 15: car0: leaves the carwash
#> 15: car1: leaves the carwash
#> 16: car2: arrives at the carwash
#> 16: car2: enters the carwash
#> simmer environment: anonymous | now: 20 | next: 21
#> { Monitor: in memory }
#> { Resource: wash | monitored: TRUE | server status: 1(2) | queue status: 0(Inf) }
#> { Source: car_initial | monitored: 1 | n_generated: 4 }
#> { Source: car | monitored: 1 | n_generated: 4 }
Covers:
From here:
This example comprises a workshop with
n
identical machines. A stream of jobs (enough to keep the machines busy) arrives. Each machine breaks down periodically. Repairs are carried out by one repairman. The repairman has other, less important tasks to perform, too. Broken machines preempt theses tasks. The repairman continues them when he is done with the machine repair. The workshop works continuously.
First of all, we load the libraries and prepare the environment.
library(simmer)
PT_MEAN <- 10.0 # Avg. processing time in minutes
PT_SIGMA <- 2.0 # Sigma of processing time
MTTF <- 300.0 # Mean time to failure in minutes
BREAK_MEAN <- 1 / MTTF # Param. for exponential distribution
REPAIR_TIME <- 30.0 # Time it takes to repair a machine in minutes
JOB_DURATION <- 30.0 # Duration of other jobs in minutes
NUM_MACHINES <- 10 # Number of machines in the machine shop
WEEKS <- 4 # Simulation time in weeks
SIM_TIME <- WEEKS * 7 * 24 * 60 # Simulation time in minutes
# setup
set.seed(42)
env <- simmer()
The make_parts
trajectory defines a machine’s operating
loop. A worker seizes the machine and starts manufacturing and counting
parts in an infinite loop. Provided we want more than one machine, we
parametrise the trajectory as a function of the machine:
make_parts <- function(machine)
trajectory() %>%
seize(machine, 1) %>%
timeout(function() rnorm(1, PT_MEAN, PT_SIGMA)) %>%
set_attribute("parts", 1, mod="+") %>%
rollback(2, Inf) # go to 'timeout' over and over
Repairman’s unimportant jobs may be modelled in the same way (without the accounting part):
Failures are high-priority arrivals, both for the machines and for the repairman. Each random generated failure will randomly select and seize (break) a machine, and will seize (call) the repairman. After the machine is repaired, both resources are released and the corresponding workers begin where they left.
machines <- paste0("machine", 1:NUM_MACHINES-1)
failure <- trajectory() %>%
select(machines, policy = "random") %>%
seize_selected(1) %>%
seize("repairman", 1) %>%
timeout(REPAIR_TIME) %>%
release("repairman", 1) %>%
release_selected(1)
The machines and their workers are appended to the simulation environment. Note that each machine, which is defined as preemptive, has space for one worker (or a failure) and no space in queue.
for (i in machines) env %>%
add_resource(i, 1, 0, preemptive = TRUE) %>%
add_generator(paste0(i, "_worker"), make_parts(i), at(0), mon = 2)
The same for the repairman, but this time the queue is infinite, since there could be any number of machines at any time waiting for repairments.
env %>%
add_resource("repairman", 1, Inf, preemptive = TRUE) %>%
add_generator("repairman_worker", other_jobs, at(0)) %>%
invisible
Finally, the failure generator is defined with
priority=1
(default: 0), and the simulation begins:
env %>%
add_generator("failure", failure,
function() rexp(1, BREAK_MEAN * NUM_MACHINES),
priority = 1) %>%
run(SIM_TIME) %>% invisible
The last value per worker from the attributes table reports the number of parts made:
aggregate(value ~ name, get_mon_attributes(env), max)
#> name value
#> 1 machine0_worker0 3250
#> 2 machine1_worker0 3303
#> 3 machine2_worker0 3241
#> 4 machine3_worker0 3336
#> 5 machine4_worker0 3399
#> 6 machine5_worker0 3423
#> 7 machine6_worker0 3304
#> 8 machine7_worker0 3181
#> 9 machine8_worker0 3253
#> 10 machine9_worker0 3169
Covers:
From here:
This example models a movie theater with one ticket counter selling tickets for three movies (next show only). People arrive at random times and try to buy a random number (1-6) of tickets for a random movie. When a movie is sold out, all people waiting to buy a ticket for that movie renege (leave the queue).
First of all, we load the libraries and prepare the environment.
library(simmer)
TICKETS <- 50 # Number of tickets per movie
SIM_TIME <- 120 # Simulate until
movies <- c("R Unchained", "Kill Process", "Pulp Implementation")
# setup
set.seed(42)
env <- simmer()
The main actor of this simulation is the moviegoer, a process that will try to buy a number of tickets for a certain movie. The logic is as follows:
This recipe is directly translated into a simmer
trajectory as follows:
get_movie <- function() movies[get_attribute(env, "movie")]
soldout_signal <- function() paste0(get_movie(), " sold out")
check_soldout <- function() get_capacity(env, get_movie()) == 0
check_tickets_available <- function()
get_server_count(env, get_movie()) > (TICKETS - 2)
moviegoer <- trajectory() %>%
# select a movie
set_attribute("movie", function() sample(3, 1)) %>%
select(get_movie) %>%
# set reneging condition
renege_if(soldout_signal) %>%
# leave immediately if the movie was already sold out
leave(check_soldout) %>%
# wait for my turn
seize("counter", 1) %>%
# buy tickets
seize_selected(
function() sample(6, 1), continue = FALSE,
reject = trajectory() %>%
timeout(0.5) %>%
release("counter", 1)
) %>%
# abort reneging condition
renege_abort() %>%
# check the tickets available
branch(
check_tickets_available, continue = TRUE,
trajectory() %>%
set_capacity_selected(0) %>%
send(soldout_signal)
) %>%
timeout(1) %>%
release("counter", 1) %>%
# watch the movie
wait()
which actually constitutes a quite interesting subset of
simmer
’s capabilities. The next step is to add the required
resources to the environment env
: the three movies and the
ticket counter.
# add movies as resources with capacity TICKETS and no queue
for (i in movies) env %>%
add_resource(i, TICKETS, 0)
# add ticket counter with capacity 1 and infinite queue
env %>% add_resource("counter", 1, Inf)
#> simmer environment: anonymous | now: 0 | next:
#> { Monitor: in memory }
#> { Resource: R Unchained | monitored: TRUE | server status: 0(50) | queue status: 0(0) }
#> { Resource: Kill Process | monitored: TRUE | server status: 0(50) | queue status: 0(0) }
#> { Resource: Pulp Implementation | monitored: TRUE | server status: 0(50) | queue status: 0(0) }
#> { Resource: counter | monitored: TRUE | server status: 0(1) | queue status: 0(Inf) }
And finally, we attach an exponential moviegoer generator to
the moviegoer
trajectory and the simulation starts:
# add a moviegoer generator and start simulation
env %>%
add_generator("moviegoer", moviegoer, function() rexp(1, 1 / 0.5), mon=2) %>%
run(SIM_TIME)
#> simmer environment: anonymous | now: 120 | next: 120.027892152333
#> { Monitor: in memory }
#> { Resource: R Unchained | monitored: TRUE | server status: 50(0) | queue status: 0(0) }
#> { Resource: Kill Process | monitored: TRUE | server status: 50(0) | queue status: 0(0) }
#> { Resource: Pulp Implementation | monitored: TRUE | server status: 49(0) | queue status: 0(0) }
#> { Resource: counter | monitored: TRUE | server status: 0(1) | queue status: 0(Inf) }
#> { Source: moviegoer | monitored: 2 | n_generated: 223 }
The analysis is performed with standard R tools:
# get the three rows with the sold out instants
sold_time <- get_mon_resources(env) %>%
subset(resource != "counter" & capacity == 0)
# get the arrivals that left at the sold out instants
# count the number of arrivals per movie
n_reneges <- get_mon_arrivals(env) %>%
subset(finished == FALSE & end_time %in% sold_time$time) %>%
merge(get_mon_attributes(env)) %>%
transform(resource = movies[value]) %>%
aggregate(value ~ resource, data=., length)
# merge the info and print
invisible(apply(merge(sold_time, n_reneges), 1, function(i) {
cat("Movie '", i["resource"], "' was sold out in ", i["time"], " minutes.\n",
" Number of people that left the queue: ", i["value"], "\n", sep="")
}))
#> Movie 'Kill Process' was sold out in 53.09917 minutes.
#> Number of people that left the queue: 9
#> Movie 'Pulp Implementation' was sold out in 51.59917 minutes.
#> Number of people that left the queue: 9
#> Movie 'R Unchained' was sold out in 33.09917 minutes.
#> Number of people that left the queue: 8
Covers:
From here:
This example models a gas station and cars that arrive for refuelling. The gas station has a limited number of fuel pumps and a fuel tank that is shared between the fuel pumps.
Vehicles arriving at the gas station first request a fuel pump from the station. Once they acquire one, they try to take the desired amount of fuel from the fuel pump. They leave when they are done.
The fuel level is reqularly monitored by a controller. When the level drops below a certain threshold, a tank truck is called for refilling the tank.
Parameters and setup:
library(simmer)
GAS_STATION_SIZE <- 200 # liters
THRESHOLD <- 10 # Threshold for calling the tank truck (in %)
FUEL_TANK_SIZE <- 50 # liters
FUEL_TANK_LEVEL <- c(5, 25) # Min/max levels of fuel tanks (in liters)
REFUELING_SPEED <- 2 # liters / second
TANK_TRUCK_TIME <- 300 # Seconds it takes the tank truck to arrive
T_INTER <- c(30, 100) # Create a car every [min, max] seconds
SIM_TIME <- 1000 # Simulation time in seconds
# setup
set.seed(42)
env <- simmer()
SimPy solves this problem using a special kind of resource called
container, which is not present in simmer
so far.
However, a container is nothing more than an embellished counter.
Therefore, we can implement this in simmer
with a global
counter (GAS_STATION_LEVEL
here), some signaling and some
care checking the counter bounds.
Let us consider just the refuelling process in the first place. A car needs to check whether there is enough fuel available to fill its tank. If not, it needs to block until the gas station is refilled.
GAS_STATION_LEVEL <- GAS_STATION_SIZE
signal <- "gas station refilled"
refuelling <- trajectory() %>%
# check if there is enough fuel available
branch(function() FUEL_TANK_SIZE - get_attribute(env, "level") > GAS_STATION_LEVEL,
continue = TRUE,
# if not, block until the signal "gas station refilled" is received
trajectory() %>%
trap(signal) %>%
wait() %>%
untrap(signal)
) %>%
# refuel
timeout(function() {
liters_required <- FUEL_TANK_SIZE - get_attribute(env, "level")
GAS_STATION_LEVEL <<- GAS_STATION_LEVEL - liters_required
return(liters_required / REFUELING_SPEED)
})
Now a car’s trajectory is straightforward. First, the starting time is saved as well as the tank level. Then, the car seizes a pump, refuels and leaves.
car <- trajectory() %>%
set_attribute(c("start", "level"), function()
c(now(env), sample(FUEL_TANK_LEVEL[1]:FUEL_TANK_LEVEL[2], 1))) %>%
log_("arriving at gas station") %>%
seize("pump", 1) %>%
# 'join()' concatenates the refuelling trajectory here
join(refuelling) %>%
release("pump", 1) %>%
log_(function()
paste0("finished refuelling in ", now(env) - get_attribute(env, "start"), " seconds"))
The tank truck takes some time to arrive. Then, the gas station is completely refilled and the signal “gas station refilled” is sent to the blocked cars.
tank_truck <- trajectory() %>%
timeout(TANK_TRUCK_TIME) %>%
log_("tank truck arriving at gas station") %>%
log_(function() {
refill <- GAS_STATION_SIZE - GAS_STATION_LEVEL
GAS_STATION_LEVEL <<- GAS_STATION_SIZE
paste0("tank truck refilling ", refill, " liters")
}) %>%
send(signal)
The controller periodically checks the GAS_STATION_LEVEL
and calls the tank truck when needed.
controller <- trajectory() %>%
branch(function() GAS_STATION_LEVEL / GAS_STATION_SIZE * 100 < THRESHOLD,
continue = TRUE,
trajectory() %>%
log_("calling the tank truck") %>%
join(tank_truck)
) %>%
timeout(10) %>%
rollback(2, Inf)
Finally, we need to add a couple of pumps, a controller worker and a car generator.
env %>%
add_resource("pump", 2) %>%
add_generator("controller", controller, at(0)) %>%
add_generator("car", car, function() sample(T_INTER[1]:T_INTER[2], 1)) %>%
run(SIM_TIME)
#> 78: car0: arriving at gas station
#> 98.5: car0: finished refuelling in 20.5 seconds
#> 172: car1: arriving at gas station
#> 190: car1: finished refuelling in 18 seconds
#> 219: car2: arriving at gas station
#> 233.5: car2: finished refuelling in 14.5 seconds
#> 295: car3: arriving at gas station
#> 314.5: car3: finished refuelling in 19.5 seconds
#> 361: car4: arriving at gas station
#> 377: car4: finished refuelling in 16 seconds
#> 410: car5: arriving at gas station
#> 442: car6: arriving at gas station
#> 498: car7: arriving at gas station
#> 532: car8: arriving at gas station
#> 595: car9: arriving at gas station
#> 627: car10: arriving at gas station
#> 698: car11: arriving at gas station
#> 742: car12: arriving at gas station
#> 807: car13: arriving at gas station
#> 854: car14: arriving at gas station
#> 952: car15: arriving at gas station
#> simmer environment: anonymous | now: 1000 | next: 1000
#> { Monitor: in memory }
#> { Resource: pump | monitored: TRUE | server status: 2(2) | queue status: 9(Inf) }
#> { Source: controller | monitored: 1 | n_generated: 1 }
#> { Source: car | monitored: 1 | n_generated: 17 }
Note though that this model has an important flaw, both in the original and in this adaptation, as discussed in #239: when two cars require more fuel than available, but the level is still above the threshold, these cars completely block the queue, and the refilling truck is never called. It could be solved by detecting this situation in the controller.