msq: Multi-Server Queue Simulation
In simEd: Simulation Education

View source: R/msq.R

msq	R Documentation

Multi-Server Queue Simulation

Description

A next-event simulation of a single-queue multiple-server service node, with extensible arrival and service processes.

Usage

msq(
  maxArrivals = Inf,
  seed = NA,
  numServers = 2,
  serverSelection = c("LRU", "LFU", "CYC", "RAN", "ORD"),
  interarrivalFcn = NULL,
  serviceFcn = NULL,
  interarrivalType = "M",
  serviceType = "M",
  maxTime = Inf,
  maxDepartures = Inf,
  maxInSystem = Inf,
  maxEventsPerSkyline = 15,
  saveAllStats = FALSE,
  saveInterarrivalTimes = FALSE,
  saveServiceTimes = FALSE,
  saveWaitTimes = FALSE,
  saveSojournTimes = FALSE,
  saveNumInQueue = FALSE,
  saveNumInSystem = FALSE,
  saveServerStatus = FALSE,
  showOutput = TRUE,
  animate = FALSE,
  showQueue = NULL,
  showSkyline = NULL,
  showSkylineSystem = FALSE,
  showSkylineQueue = FALSE,
  showSkylineServer = FALSE,
  showTitle = TRUE,
  showProgress = TRUE,
  plotQueueFcn = defaultPlotMSQ,
  plotSkylineFcn = defaultPlotSkyline,
  jobImage = NA,
  plotDelay = NA,
  respectLayout = FALSE
)

Arguments

`maxArrivals`	maximum number of customer arrivals allowed to enter the system
`seed`	initial seed to the random number generator (NA uses current state of random number generator; NULL seeds using system clock)
`numServers`	Number of servers to simulation (an integer between 1 and 24)
`serverSelection`	Algorithm to use for selecting among idle servers (default is "LRU")
`interarrivalFcn`	Function for generating interarrival times for queue simulation. Default value (`NA`) will result in use of default interarrival function based on `interarrivalType`. See examples.
`serviceFcn`	Function for generating service times for queue simulation. Default value (`NA`) will result in use of default service function based on `serviceType`. See examples.
`interarrivalType`	string representation of desired interarrival process. Options are "M" – exponential with rate 1; "G" – uniform(0,2), having mean 1; and "D" – deterministic with constant value 1. Default is "M".
`serviceType`	string representation of desired service process . Options are "M" – exponential with rate 10/9; "G" – uniform(0, 1.8), having mean 9/10; and "D" – deterministic with constant value 9/10. Default is "M".
`maxTime`	maximum time to simulate
`maxDepartures`	maximum number of customer departures to process
`maxInSystem`	maximum number of customers that the system can hold (server(s) plus queue). Infinite by default.
`maxEventsPerSkyline`	maximum number of events viewable at a time in the skyline plot. A large value for this parameter may result in plotting delays. This parameter does not impact the final plotting, which will show all end-of-simulation results.
`saveAllStats`	if `TRUE`, returns all vectors of statistics (see below) collected by the simulation
`saveInterarrivalTimes`	if `TRUE`, returns a vector of all interarrival times generated
`saveServiceTimes`	if `TRUE`, returns a vector of all service times generated
`saveWaitTimes`	if `TRUE`, returns a vector of all wait times (in the queue) generated
`saveSojournTimes`	if `TRUE`, returns a vector of all sojourn times (time spent in the system) generated
`saveNumInQueue`	if `TRUE`, returns a vector of times and a vector of counts for whenever the number in the queue changes
`saveNumInSystem`	if `TRUE`, returns a vector of times and a vector of counts for whenever the number in the system changes
`saveServerStatus`	if `TRUE`, returns a vector of times and a vector of server status (0:idle, 1:busy) for whenever the status changes
`showOutput`	if `TRUE`, displays summary statistics upon completion
`animate`	If FALSE, no animation will be shown.
`showQueue`	if TRUE, displays a visualization of the queue
`showSkyline`	If `NULL` (default), defers to each individual showSkyline... parameter below; otherwise, supersedes individual showSkyline... parameter values. If `TRUE`, displays full skyline plot; `FALSE` suppresses skyline plot. Can alternatively be specified using chmod-like octal component specification: use 1, 2, 4 for system, queue, and server respectively, summing to indicate desired combination (e.g., 7 for all). Can also be specified as a binary vector (e.g., c(1,1,1) for all).
`showSkylineSystem`	logical; if `TRUE`, includes number in system as part of skyline plot. Value for `showSkyline` supersedes this parameter's value.
`showSkylineQueue`	logical; if `TRUE`, includes number in queue as part of skyline plot. Value for `showSkyline` supersedes this parameter's value.
`showSkylineServer`	logical; if `TRUE`, includes number in server as part of skyline plot. Value for `showSkyline` supersedes this parameter's value.
`showTitle`	if `TRUE`, display title at the top of the main plot
`showProgress`	if TRUE, displays a progress bar on screen during no-animation execution
`plotQueueFcn`	Plotting function to display Queue visualization. By default, this is provided by `defaultPlotSSQ`. Please refer to the corresponding help for more details about required arguments.
`plotSkylineFcn`	Plotting function to display Skyline visualization. By default, this is provided by `defaultPlotSkyline`. Please refer to the corresponding help for more details about required arguments.
`jobImage`	a vector of URLs/local addresses of images to use as jobs. Requires package `'magick'`.
`plotDelay`	a positive numeric value indicating seconds between plots. A value of -1 enters 'interactive' mode, where the state will pause for user input at each step. A value of 0 will display only the final end-of-simulation plot.
`respectLayout`	If TRUE, plot layout (i.e., par, device, etc.) settings will be respected. Not recommended except for specialized use.

Details

Implements a next-event implementation of a single-queue multiple-server queue simulation.

The seed parameter can take one of three valid argument types:

NA (default), which will use the current state of the random number generator without explicitly setting a new seed (see examples);
a positive integer, which will be used as the initial seed passed in an explicit call to set.seed; or
NULL, which will be passed in an explicit call to to set.seed, thereby setting the initial seed using the system clock.

The server selection mechanism can be chosen from among five options, with "LRU" being the default:

"LRU" (least recently used): from among the currently available (idle) servers, selects the server who has been idle longest.
"LFU" (least frequently used): from among the currently available servers, selects the server having the lowest computed utilization.
"CYC" (cyclic): selects a server in a cyclic manner; i.e, indexing the servers 1, 2, \ldots, numServers and incrementing cyclically, starts from one greater than the index of the most recently engaged server and selects the first idle server encountered.
"RAN" (random): selects a server at random from among the currently available servers.
"ORD" (in order): indexing the servers 1, 2, \ldots, numServers, selects the idle server having the lowest index.

Value

The function returns a list containing:

the number of arrivals to the system (customerArrivals),
the number of customers processed (customerDepartures),
the ending time of the simulation (simulationEndTime),
average wait time in the queue (avgWait),
average time in the system (avgSojourn),
average number in the system (avgNumInSystem),
average number in the queue (avgNumInQueue), and
server utilization (utilization).

of the queue as computed by the simulation. When requested via the “save...” parameters, the list may also contain:

a vector of interarrival times (interarrivalTimes),
a vector of wait times (waitTimes),
a vector of service times (serviceTimes),
a vector of sojourn times (sojournTimes),
two vectors (time and count) noting changes to number in the system (numInSystemT, numInSystemN),
two vectors (time and count) noting changes to number in the queue (numInQueueT, numInQueueN), and
two vectors (time and status) noting changes to server status (serverStatusT, serverStatusN).

Author(s)

Barry Lawson (blawson@bates.edu),
Larry Leemis (leemis@math.wm.edu),
Vadim Kudlay (vkudlay@nvidia.com)

Examples

 # process 100 arrivals, R-provided seed (via NULL seed), default 2 servers
 msq(100, NULL)
 # process 100 arrivals, seed 8675309, 3 servers, LFU server selection
 msq(100, 8675309, 3, 'LFU')

 msq(maxArrivals = 100, seed = 8675309)
 msq(maxTime = 100, seed = 8675309)

 ############################################################################
 # example to show use of seed = NA (default) to rely on current state of generator
 output1 <- msq(200, 8675309, showOutput = FALSE, saveAllStats = TRUE)
 output2 <- msq(300,          showOutput = FALSE, saveAllStats = TRUE)
 set.seed(8675309)
 output3 <- msq(200,          showOutput = FALSE, saveAllStats = TRUE)
 output4 <- msq(300,          showOutput = FALSE, saveAllStats = TRUE)
 sum(output1$sojournTimes != output3$sojournTimes) # should be zero
 sum(output2$sojournTimes != output4$sojournTimes) # should be zero

 ############################################################################
 # use same service function for (default) two servers
 myArrFcn <- function() { vexp(1, rate = 1/4, stream = 1) }               # mean is 4
 mySvcFcn <- function() { vgamma(1, shape = 1, rate = 0.3, stream = 2) }  # mean is 3.3
 output <- msq(maxArrivals = 100, interarrivalFcn = myArrFcn,
     serviceFcn = mySvcFcn, saveAllStats = TRUE)
 mean(output$interarrivalTimes)
 mean(output$serviceTimes)

 ############################################################################
 # use different service function for (default) two servers
 myArrFcn  <- function() { vexp(1, rate = 1/4, stream = 1) }                # mean is 4
 mySvcFcn1 <- function() { vgamma(1, shape = 3, scale = 1.1, stream = 2) }  # mean is 3.3
 mySvcFcn2 <- function() { vgamma(1, shape = 3, scale = 1.2, stream = 3) }  # mean is 3.6
 output <- msq(maxArrivals = 100, interarrivalFcn = myArrFcn,
     serviceFcn = list(mySvcFcn1, mySvcFcn2), saveAllStats = TRUE)
 mean(output$interarrivalTimes)
 meanTPS(output$numInQueueT, output$numInQueueN)  # compute time-averaged num in queue
 mean(output$serviceTimesPerServer[[1]])  # compute avg service time for server 1
 mean(output$serviceTimesPerServer[[2]])  # compute avg service time for server 2
 meanTPS(output$serverStatusT[[1]], output$serverStatusN[[1]])  # compute server 1 utilization
 meanTPS(output$serverStatusT[[2]], output$serverStatusN[[2]])  # compute server 2 utilization

 ############################################################################
 # example to show use of (simple) trace data for arrivals and service times,
 # allowing for reuse of trace data times
 smallQueueTrace <- list()
 smallQueueTrace$arrivalTimes <- c(15, 47, 71, 111, 123, 152, 166, 226, 310, 320)
 smallQueueTrace$serviceTimes <- c(43, 36, 34,  30,  38,  40,  31,  29,  36,  30)

 interarrivalTimes <- NULL
 serviceTimes      <- NULL

 getInterarr <- function()
 {
     if (length(interarrivalTimes) == 0) {
           interarrivalTimes <<- c(smallQueueTrace$arrivalTimes[1],
                                  diff(smallQueueTrace$arrivalTimes))
     }
     nextInterarr <- interarrivalTimes[1]
     interarrivalTimes <<- interarrivalTimes[-1]  # remove 1st element globally
     return(nextInterarr)
 }

 getService <- function()
 {
     if (length(serviceTimes) == 0) {
         serviceTimes <<- smallQueueTrace$serviceTimes
     }
     nextService <- serviceTimes[1]
     serviceTimes <<- serviceTimes[-1]  # remove 1st element globally
     return(nextService)
 }

 output <- msq(maxArrivals = 100, numServers = 2, interarrivalFcn = getInterarr,
               serviceFcn = getService, saveAllStats = TRUE)
 mean(output$interarrivalTimes)
 mean(output$serviceTimes)
 mean(output$serviceTimesPerServer[[1]])  # compute avg service time for server 1
 mean(output$serviceTimesPerServer[[2]])  # compute avg service time for server 2

 ############################################################################
 # Testing with visualization

 # Visualizing msq with a set seed, infinite queue capacity, 10 arrivals,
 # and showing queue (default) and skyline for all 3 attributes
 msq(seed = 1234, numServers = 5, maxArrivals = 10, showSkyline = 7, 
     plotDelay = 0.1)

 # Same simulation as above but using default interactive mode
 if (interactive()) {
   msq(seed = 1234, numServers = 5, maxArrivals = 10, showSkyline = 7)
 }

 # Visualizing msq with a set seed, finite queue capacity, 20 arrivals,
 # and showing queue (default) and skyline for all 3 attributes
 msq(seed = 1234, numServers = 5, maxArrivals = 25, showSkyline = 7,
     maxInSystem = 5, plotDelay = 0)

 # Using default distributions to simulate an M/G/2 queue
 msq(seed = 1234, maxDepartures = 10, 
     interarrivalType = "M", serviceType = "G", plotDelay = 0)

simEd documentation built on Nov. 27, 2023, 1:07 a.m.