Products/FAQ : Introduction to CSIM

In CSIM, a CSIM process is a procedure which executes the create statement. Thus, a typical CSIM process has the form:

void customer() // defines customer
{
create("cust");
// other statements
}

A CSIM process is started (invoked) by another process; it is called using the usual procedure call:

void otherProcess()
{
create("other");
// other statements
customer(); // starts customer process
// other statements
}

CSIM processes are different from normal procedures in several important ways:

The flow of control (the "semantics") of invoking processes is illustrated in the following example:

void otherProcess() // the starting process
{
long i;
create("other");
for(i = 0; i < 10; i++) // invoke 10 instances of cust
customer();
hold(25.0);
// other statements
}
void customer() // the started process
{
create("customer"); // <--- Focus of discussion
// other statements
}

Understanding the operation of the create statement in the customer() process is critical to understanding the operation of CSIM processes. When this create statement executes, the following actions take place:

In this example, the "otherProcess" process invokes 10 instances of the "customer" process. At the end of the "for" loop, there are 10 instances of "customer", each Ready to enter the Computing state. When "otherProcess" executes the hold(25.0) statement, it is suspended and the first "customer" process enters the Computing state and continues computing until it suspends itself. At this point, the second "customer" process starts computing, and so on.

The hold(25.0) statement really both suspends "otherProcess" and schedules it to enter the Ready state when 25.0 units of simulated time have passed. It should be noted that simulated time passes only as the result of processes executing hold statements. It should also be noted that "computing" take no simulated time; in other words, computing is "free", in the sense that computing time does not require simulated time.

The remainder of this note discusses the simulated resources and other structures which are used to control the processes of a model and the interactions between these processes.

Resources

In many systems, processes must visit and use (occupy) a succession of resources. CSIM offers two kinds of resources: facilities and storages. A facility consists of a single queue (for waiting processes) and one or more servers. Facilities are used to represent resources where entities (processes) "use" servers "one-at-a-time". The operations which are applied to facilities by processes include:

In addition to the queue and server(s), a facility also has provisions for collecting data on the delays associated with gaining access to a server and on using the servers. This data collection is automatic; a report summarizing these collected data can be produced at any time during the execution of the model.

A storage consists of a queue and a pool of storage units (sometimes called tokens). A process can allocate one or more storage units; if there is not a sufficient number of units to satisfy an allocation request, the process is suspended and placed in the queue. When other processes have deallocated their storage units, queued processes are given units to satisfy their requests. As with facilities, there are data which summarize the delays for and use of these storage units.

Process Interactions

CSIM provides two structures to facilitate and control interactions and communications between different processes. These structures are:

An event can be in either the OCCURRED state or the NOT_OCCURRED state. A process can wait for a specified event to "occur". Another process can set an event, causing it to be placed in the OCCURRED state and allowing all of the waiting processes to resume (enter the Ready state).

A mailbox is a place where processes can exchange information (messages). One process can send a message to a mailbox. Another process can attempt to receive a message from a mailbox; if a message is "in" the mailbox, the receiving process gets that message and continues computing; if there are no messages in the mailbox, each receiving process must wait until a message is sent to that mailbox.

An Example

As an example, consider an assembly line, where TV sets pass through an assembly station and then are inspected. About 20% of the inspected sets fail and are sent back to the assembly station for rework. If a set fails inspection a second time, it is scrapped. The parameters of the model are the arrival rate of sets to this station, the time required to assemble or rework a set, the number of workers (servers) at the assembly station, the number of inspectors and the inspection failure rate. The goal is to determine the number of workers and inspectors required to insure that this station is not a "bottleneck" in the operation of the factory.

A complete CSIM model of this system follows as Figure 1; the report output for this model appears in Figure 2. The model is a C++ program which utilizes the classes, methods and procedures of the CSIM library.

// Model TV Set Assembly and Inspection Stations

#include "cpp.h"
const double simTime = 8*60.0; // parameters, constants
const long numberWorkers = 5;
const long numberInspectors = 2;
const double arrTime = 2.0;
const double assemblyTime = 3.0;
const double reworkTime = 1.5;
const double inspectTime = 2.0;
const double failRate = 0.20;
facility initialPrep("initial"); // facilities
facility_ms assembly("assmbly", numberWorkers);
facility_ms inspect("inspect", numberInspectors);
facility finalPrep("final");
facility scrapPrep("scrap");
FILE *fp;
void generate(); // prototypes
void TVSet();

extern "C" void sim() // main process
{
fp = fopen("xxx.out", "w");
set_output_file(fp);
create("sim");
generate(); // start generate process
hold(simTime); // lengthof run
report(); // producereport
}

void generate() // generator process
{
create("gen");
while(1) {
TVSet(); // start TV Set process
hold(exponential(arrTime)); // timebetween TV Sets
}
}

void TVSet() // TV Set process
{
long inspectCt = 0;
double stationTime;
long exitCond = 0;
create("TVSet");
initialPrep.use(0.5); // prep station
stationTime = assemblyTime;
do {
assembly.use(exponential(stationTime)); // assembly station
inspect.use(exponential(inspectTime)); // inspect station
inspectCt++;
if(bernoulli(failRate) == 0) { // outcome of inspection
finalPrep.use(0.5); // passed -> final prep
exitCond = 1;
}
else {
if(inspectCt < 2) {
stationTime = reworkTime; // failed 1st, rework
}
else {
scrapPrep.use(0.5); // failed 2nd, -> scrap
exitCond = 1;
}
}
} while(exitCond == 0);
}

Figure 1: Listing of Example Model (TV Sets)

CSIM Simulation Report (Version 19 for MS Visual C++)

Tue Apr 08 13:25:46 1997

Ending simulation time: 480.000
Elapsed simulation time: 480.000
CPU time used (seconds): 0.160

Figure 2: Report Output for TV Set Example

The model has two kinds of processes (in addition to the initial "sim" process): the "generate" process injects TVSet processes into the model at varying intervals which model the arrivals of TV Sets at this station. The probability distribution of these interarrival intervals is a negative exponential distribution with mean as specified by the arrTime constant. The TVSet process models the behavior of a TV set as it progresses through the set of stations. The facilities (representing the stations of the system) are declared and constructed a global, static objects. Each TVSet process uses (visits), in turn, the initialPrep station, the assembly station and the inspection station. The times spent at each station are all sampled from negative exponential probability distributions, with means specified by the appropriate constants. CSIM provides a variety of probability driven random number functions, so that it is likely that an appropriate function can be found for any modeling situation.

The outcome of the inspection is determined by a Bernoulli function; the parameter is the mean failure rate. Thus, a successful outcome of the Bernoulli function is a failed TV set and an unsuccessful outcome of the Bernoulli function is a passed TV set. Passed sets go to the finalPrep station and then exit the model. Failed sets can return to the assembly station for a rework cycle, but only one time. Thus, a unit failing for the second time is sent to the scrapPrep station and then exits.

The report output shows that with five assembly workers and two inspectors, the model predicts "acceptable" performance. Other runs of the model, for example with one inspector, show that the number of inspectors can be critical to successful operation. With one inspector, the response time at the inspector station climbs to about 20 minutes per TV set.

Other Features

CSIM includes many features which were not described above. For example, there are extensive capabilities for collecting specialized data, which are needed to gain insight into specific aspects of the operation of the model. Both real-valued items, such as response times, and time-based, integer-valued items (such as queue lengths) can be collected. Included in these data gathering features are special mechanisms for producing confidence intervals for output values. There is also a "automatic run-length control" feature, which lets a model execute until desired levels of statistical accuracy for specified output values are achieved.

All of the data collected at facilities, storages, tables (for real valued data) and qtables (for time-based, integer-valued data) are accessible via a group of inspector functions. These can be used to produce output reports tailored to specific reporting needs.

There are capabilities for testing the status of the simulated resources and for modifying the characteristics of these resources. These can be used to produce very accurate representations of many different kinds of system resources.

All of the features and capabilities of all of the classes, methods and procedures in the CSIM library are described in the CSIM User's Guide.

Summary

This Primer has described some of the features and capabilities of the CSIM simulation software toolkit. The Primer has focused on CSIM processes, as these are critical to implementing robust models of complex systems. A fairly detailed example illustrated the use of processes and facilities to model a section of a TV set assembly line. The interested reader is referred to the CSIM User's Guides and the CSIM Getting Started manuals for additional information.