)
The Social Networks package comes with an example about doing this with the Network field.
If you use a Continuous2D, you'll want to make that your discretization significantly larger than the object you're storing. If you just have a single object, we suggest making the discretization as large or larger than the entire region of the field (so it's just one big cell).
The most common need is probably to put a map in the background of the screen. For this, your SimplePortrayal2D could be an ImagePortrayal2D. Make sure your field's width and height are of the proportions you want. Then add the object to the exact center of the field you've created. Now attach an ImagePortrayal2D along these lines:
myContinuousPortrayal2D.setField(theContinuousField);
myContinuousPortrayal2D.setPortrayalForAll(new ImagePortrayal2D(
new ImageIcon(getClass().getResource("MyPicture.gif")).getImage()),theContinuousField.width);
myDisplay.attach(myContinuousPortrayal2D, "Overlay");
If the image is taller than wide, change theContinuousField.width to theContinuousField.height. If you're filling the entire background, remember to set Display2D's backdrop to null so it doesn't bother drawing it.
FieldPortrayal2D overlay = new FieldPortrayal2D()
{
Font font = new Font("SansSerif", 0, 18); // keep it around for efficiency
public void draw(Object object, Graphics2D graphics, DrawInfo2D info)
{
String s = "";
if (state !=null)
s = state.schedule.getTimestamp("Before Simulation", "All Done!");
graphics.setColor(Color.blue);
graphics.drawString(s, (int)info.clip.x + 10,
(int)(info.clip.y + 10 + font.getStringBounds(s,graphics.getFontRenderContext()).getHeight()));
}
};
You don't need to set the field of this FieldPortrayal2D -- just leave it empty. All you need to do is attach it to your Display and you're on your way.
Note that this method locks to the top-left corner of the clip region. If you have scrolled out a lot, the clip region will be smaller than the window and located in the center of it. If you want to always draw in the top-left corner of the screen, you need to tell the Display2D to stop clipping. Then the clip region will always be the scroll rectangle. However if any of your portrayals spill outside of the clip region, their unclipped drawing will now be visible.
If you're filling the entire background with this method, remember to set Display2D's backdrop to null so it doesn't bother drawing it.
Another way to draw a scale-independent image in the background is to use a TexturePaint wrapped around your image, and use that as the "Paint" to draw the Display2D's backdrop. You could, for example, add this to your setupPortrayals method:
Image i = new ImageIcon(getClass().getResource("MyPicture.gif")).getImage();
BufferedImage b = display.getGraphicsConfiguration().createCompatibleImage(i.getWidth(null), i.getHeight(null));
Graphics g = b.getGraphics();
g.drawImage(i,0,0,i.getWidth(null),i.getHeight(null),null);
g.dispose();
display.setBackdrop(new TexturePaint(b, new Rectangle(0,0,i.getWidth(null),i.getHeight(null))));
This will tile the image into the scrolled space. Unfortunately, large tiled images are buggy on certain platforms (such as OS X java 1.3.x). So we don't recommend it.
As an example, we will convert Conway's Game of Life from Tutorials 1 and 2 into an applet.
The simulation.classes file, located in the sim/display directory, specifies which files will appear in the pop-up list when the user chooses "New Simulation...". Edit this file to include only those simulations you wish the user to have access to. If you put the word 'ONLY' on a line all by itself, the user will not be able to type in a simulation classname by hand. If the file consists solely of the world 'ONLY', then the "New Simulation..." menu will be grayed out and the user won't be able to make a new simulation at all.
jar cvf mason.jar masonThis dumps the entire mason directory, java files and all, into a jar file called mason.jar. You don't need all those files -- you just need:
To extract just these files and no others from the MASON directory and stuff them into a jar file, the easiest approach is (in UNIX or on a Mac) to go into the mason directory and type
make jar
This is still a big file for users to download. If you really want to shrink this, you'll need to identify which classes and related files have to be extracted. Generally speaking the only images that need to be extracted for MASON to operate are the PNG files in the display and util/gui directories. The HTML files you need are probably just those in the application directories you wish to make available. And you'll need the simulation.classes file. As to the class files: generally you're safe if you just grab all the class files in the mason directory EXCEPT for various sim/app/... applications you don't wish to include. Certain tutorials have big pictures, and you'll save a lot of space by tossing those tutorials out. If you're not doing Java3D, you can toss out the files in the portrayal3d and display3d directories as well.
Your mason.jar file should be located on the website right next to this HTML file. An example is shown on the MASON home page.
If you want to fire up MASON directly on the simulation of your choice, you'll need to change an applet parameter called Simulation in your HTML file. Its value is by default sim.display.Console, but you can change it to a simulation class, for example, sim.app.heatbugs.HeatBugsWithUI
You can also change what is placed on the button by changing the Name parameter. Note that these parameters have to be changed in several locations in our version of the file, because different web browsers handle HTML in subtly different ways.
Not all operating systems run Java 1.5. For example, most web browsers on OS X run Java 1.4.2. If you compile your applet class files using Java 1.5, by default it will compile them to a bytecode version which only runs on Java 1.5 virtual machines, even if you don't use any new Java 1.5 features. This means your applet won't run on various operating systems.
You can compile for an earlier bytecode version by changing the target of your virtual machine. It's probably safe to compile to 1.3 as your target. If you're compiling using javac on the command lline, this can be as easy as saying javac -target 1.3 ... You can do the same thing with jikes, like this: jikes -target 1.3 ...Œ
Note that parallelization uses threads and is subject to race conditions. If your parallel actions perform writes or access the random number generator, you'll need to synchronize on the schedule. Remember not to hold the lock on the schedule for long!
The MASON HeatBugs example shows a variety of approaches to improving the efficiency of the HeatBugs Diffuser, up to and including parallelizing it for multiple processors using ParallelSequence.
Generally speaking there are two kinds of Asynchronous Steppable operations: fast one-shot operations and slow or inifinite-loop operations. Fast one-shot operations can simply fire off in their own thread and do their thing, assuming they're fast enough that the user is willing to wait for them to finish after he has pressed the Stop or Pause buttons or is checkpointing out the simulation. Slow or infinite-loop operations must be able to respond to requests to be paused, resumed, and stopped (killed).
We expect that the use of Asynchronous Steppables should be exceedingly rare. Indeed, adding them to MASON required some retooling and we do not know if it works properly: it's essentially untested at this point. Read the AsynchronousSteppable documentation very closely.
MASON provides WeakSteppable, a wrapper for Steppables which holds onto them as WeakReferences. It also maintains weak Stoppables as well (though that may change). Keep in mind that just because the WeakReference is being held onto no one but the Schedule doesn't mean the Schedule will let go of it: this only occurs when Java needs to reclaim the WeakReference, perhaps during garbage collection.
When the WeakSteppable lets go of its weak reference, the WeakSteppable is still scheduled in the Schedule. It'll just do nothing when the Schedule gets around to stepping it. However, the time will advance to the scheduled point.
Schedule doesn't at present have a way to delete things except when their time is called. The reason is because Schedule uses a data structure called a binary heap to work efficiently. Heaps are good for inserting arbitrary objects and then extracting them in sorted order (in our case, sorted order means order of scheduled time). However, other arbitrary deletion of objects from binary heaps is highly inefficient. Instead, the better way to "delete" an object is to "stop" it and forget about it. It'll drop out of the Schedule in due time. How you "stop" an agent depends on whether or not it's a one-shot agent or a repeating agent, which we discuss next.
One-shot agents. Agents can be scheduled on the Schedule in two ways: one-shot agents and repeating agents. When an agent is scheduled just once, it sits in the Schedule until its time has come. At that point the agent is deleted from the Schedule and its step() method is called. If you're writing a one-shot agent, and you want your agent out of the Schedule, all you have to do is tell your agent to ignore its sole step() method and wait.
But perhaps this is inconvenient: you might want to reuse the agent and its step() method -- perhaps to reschedule it at a future time instead. Having it ignore future step() calls isn't feasible in that case. Then instead you can wrap it in another Steppable which can be stopped. We provide such an object: sim.engine.TentativeStep. When its step() is called, this object will in turn call step() on its subsidiary Steppable unless stop() has been called, at which time it just forgets about the subsidiary object (sets it to null). You can wrap your Steppable in various TentativeSteps and schedule each of them independently on the Schedule, then stop() various ones at your leisure.
Repeating agents. A repeating-scheduled agent has actually just been wrapped in a private 'Repeat' Steppable. When the step() method is called on the 'repeat' object, it simply calls step() on its subsidiary Steppable (your agent), then it reschedules itself in the Schedule for a later date. This 'repeat' object is also a Stoppable: when its stop() method is called, then the subsidiary Steppable is set to null and the 'Repeat' object ceases to function -- no later subsidiar step(), no later rescheduling.
This Stoppable is handed to you when you call scheduleRepeating(...): to delete a repeating object from the Schedule, you merely need to call stop() on the Stoppable and forget about the agent.
SparseGrid2D, SparseGrid3D, Continuous2D, Continuous3D. Use the remove(obj) method declared in these classes' superclass, SparseField. Under no circumstances should you manually remove objects from the Bags of these classes -- there's a lot of unhooking that must be done in different places. Let remove(obj) do the job for you. You can also call removeObjectsAtLocation(location) (and variations thereof) which removes and returns all objects at a given location. And you can call clear() to delete all objects in the entire field.
ObjectGrid2D, ObjectGrid3D. You remove elements from these objects by simply deleting the elements from their underlying arrays. The clear() function will delete all elements for you.
Network. Use removeEdge(edge) to delete an edge from the graph, and removeNode(node) to delete a node from the graph. Once an edge has been removed from the graph, you can reuse it by adding it to another graph, but you cannot change the nodes it is connected to (they will be added to the other graph as well if not already present). You can also delete all nodes and edges from the graph with clear().
IntGrid2D, IntGrid3D, DoubleGrid2D, DoubleGrid3D. There are no removal functions in these fields as there is nothing to "remove".
It's often the case that you might want to have a numerical property with minimum and maximum bounds. In this case you might want the user to be able to enter values with a slider rather than having to enter them in a text field (which is more cumbersome for real-time control).
It's easy to do. Let's say that your property looks like this:
Let's say you want to specify that the minimum of this property is 4 and the maximum is 18.3. What you need to do is add a new method called domMyProperty() like this.
The dom marker (which is special to MASON) tells MASON that your numerical property has a domain bounded by a minimum and maximum. MASON then will use a slider to allow the user to set it to values within that domain. You can do this with any integer (byte/short/int/long) or floating-point (float/double) property. For integer properties you should specify the minimum and maximum as longs:
Not the use of L to denote a long. This is important; otherwise Java will think the Interval is defined in terms of doubles.
If your integer (byte/short/int/long) property is defined in terms of values 0...n, you also have the option of having MASON provide a pop-up list of n names, each of which represents one of those values. For example, for value 0 you might want the pop-up list to display "North", and for value 1 you might want it to display "East", etc. Then when the user picks "East", your property is set to 1. This can be done by providing an array in this fashion:
Use Java 1.3.1 on MacOS X 1.3.1 is hardware accelerated. 1.4.2 and 1.5 are not! They are exceptionally slow at drawing 2D, and you will be shocked to find how much faster 1.3.1 is (except on Intel Macs, where at present it is extremely slow). However if you need to do 3D visualization, you have no choice but to use 1.4.2 or 1.5. MASON's mason.command script launches with the default Java (1.4.2 or 1.5), while mason.1.3.1.command launches with Java 1.3.1.
Use a SparseGrid rather than an ObjectGrid ObjectGrid portrayals generally work by going through the entire ObjectGrid and drawing each and every grid cell, whether it contains an object or not. SparseGrid just goes through its collection of objects and draws each one in turn. Whenever your environment is sparse (you have a large grid but rather few objects inhabiting it), you might try SparseGrid. The trade-off is that SparseGrid stores its objects' locations in a hash table, and so looking up or changing the location of an object requires one or several hashes, plus some wandering through arrays. This is typically an O(1) operation but it incurs a significant overhead, so if you have a lot of objects -- your environment isn't very sparse -- then SparseGrid may not buy you anything. Try and see.
Avoid ObjectGrid3D and ValueGrid3D Java3D is slow. Drawing large, non-sparse 3D grids of objects in Java3D is slow to the point of painful, and requires an enormous amount of additional memory. Avoid it if possible.
Avoid Java Graphics2D in Custom Simple Portrayals Use Java AWT graphics whenever possible: Graphics2D is generally much slower.
Zoom In MASON's Display2D only draws that portion of the fields which are actually shown on-screen; the rest is carefully ignored (exception: NetworkPortrayal2D). If you show less of the field, fewer objects are drawn.
Declare an Immutable Field If you have a field which never changes, you can declare this in certain "Fast" field portrayals (see below), and they will instead convert it into an image and just re-splat the image each time. The field portrayals which can do this: FastObjectGridPortrayal2D, FastValueGridPortrayal2D, FastHexaObjectGridPortrayal2D, FastHexaValueGridPortrayal2D.
Use a Faster Field Portrayal In several situations MASON has both regular but flexible and fast but limited field portrayals for 2D and 3D fields. All of the "fast" versions of the field portrayals draw their fields as simple grids of rectangles rather than other prettier representations. The idea behind the "fast" versions is that they draw these grids very rapidly using Java tricks rather than individually calling up simple portrayals to draw each grid region. Typically these portrayals represent elements simply as colored rectangles, bypassing simple portrayals entirely. In most cases they can draw these rectangles in two ways which the user can choose in the Options panel.
Look further down for a list of fast field portrayals.
To draw their colored rectangles, the fast field portrayals require a ColorMap which maps numerical (double) values to color representations. For example, to translate values of 0.0 through 100.0 to range smoothly from red to blue, (values outside those ranges get thresholded to red or blue respectively), you might write:
You can also have per-integer colors too, if your values range from 0...n. To set the three integers 0, 1, 2 to three different colors, you could write:
You can make SimpleColorMaps which do both things at the same time as well. See the SimpleColorMap documentation. Or you can make your own ColorMap that does something more complex (like a nonlinear translation in colors).
What about ObjectGrid2D, which doesn't have numerical values, but has actual Objects? The "fast" ObjectGrid2D portrayals, first translate the Objects into numbers, then run the numbers into the ColorMap, using the Portrayals' built-in doubleValue(...) method. The default version of the method works like this: if the object is null, 0.0 is returned. If the object is a java.util.Number or is sim.util.Valuable, then obj.doubleValue() is returned. Else 1.0 is returned. This probably works for most situations you have, but you're welcome to override it to provide more specific values.
Speed hint #1 for ColorMaps. If a large portion of your map is going to be the same color, consider making it 100% transparent (new Color(0,0,0,0)). Just set the backdrop color of the Display instead and let it bleed through. If a grid cell is completely transparent, MASON avoids drawing it all, resulting in a major speed boost.
Speed hint #2 for ColorMaps. If you're running on Linux or Windows, semi-transparent colors are expensive to draw. Use opaque or fully transparent. On OS X semi-transparency is quite fast.
Here's a chart of fields and their regular/fast portrayals (as of MASON 11).
How to Add a Slider or Pop-Up Menu to an Inspected Property
By default inspected properties take one of four forms:
public void setMyProperty(double val) { ... }
public double getMyProperty() { ... }
public void setMyProperty(double val) { ... }
public double getMyProperty() { ... }
public Object domMyProperty() { return new sim.util.Interval(4.0, 18.3); }
public void setMyOtherProperty(short val) { ... }
public short getMyOtherProperty() { ... }
public Object domMyOtherProperty() { return new sim.util.Interval(7L, 15L); }
public void setYetAnotherProperty(int val) { ... }
public int getYetAnotherProperty() { ... }
public Object domYetAnotherProperty() { return new String[] { "North", "East", "South", "West" }; }
How to Draw Faster
portrayal.setMap(new sim.util.gui.SimpleColorMap(0.0, 100.0, Color.red, Color.blue));
portrayal.setMap(new sim.util.gui.SimpleColorMap(new Color[] { Color.red, Color.blue, Color.green }));
| Field | Layout | Regular Field Portrayal | Fast Field Portrayal |
| 2D Field Portrayals | |||
| IntGrid2D or DoubleGrid2D | Rectangular | ValueGridPortrayal2D | FastValueGridPortrayal2D |
| IntGrid2D or DoubleGrid2D | Hexagonal | HexaValueGridPortrayal2D | FastHexaValueGridPortrayal2D |
| ObjectGrid2D | Rectangular | ObjectGridPortrayal2D | FastObjectGridPortrayal2D |
| ObjectGrid2D | Hexagonal | HexaObjectGridPortrayal2D | FastHexaObjectGridPortrayal2D |
| SparseGrid2D | Rectangular | SparseGridPortrayal2D | |
| SparseGrid2D | Hexagonal | HexaSparseGridPortrayal2D | |
| Continuous2D | ContinuousPortrayal2D | ||
| Network | NetworkPortrayal2D | ||
| 3D Field Portrayals | |||
| IntGrid3D or DoubleGrid3D | Rectangular | ValueGridPortrayal3D | |
| SparseGrid3D | Rectangular | SparseGridPortrayal3D | |
| ObjectGrid3D or ObjectGrid2D | Rectangular | ObjectGridPortrayal3D | |
| IntGrid2D or DoubleGrid2D | Rectangular | ValueGridPortrayal3D | ValueGrid2DPortrayal3D |
| SparseGrid2D | Rectangular | SparseGridPortrayal3D or SparseGrid2DPortrayal3D |
|
| Continuous2D or Continuous3D | ContinuousPortrayal3D | ||
| Network | NetworkPortrayal3D | ||
Scaling and translation of field portrayals usually occurs when you need to adjust different portrayals so that they lie properly on top of one another. Translation also occurs when you wish to center the origin. At present MASON has no facilities for flipping the Y axis so that it goes up rather than down. Instead, you may need to adjust your model data's arrangement.
In any of the following cases, you may also find it helpful to eliminate clipping of the display region:
display.setClipping(false);Translating By default, the origin of a Field Portrayal is in the top-left corner of the screen. If you wish to center the origin on-screen, for example, you need to translate it to the middle of the display area. This can be done by attaching the portrayal to the display not in the usual way, but like this:
display.attach(portrayal, "Portrayal Text Goes Here", display.insideDisplay.width / 2.0,
display.insideDisplay.height / 2.0, true);
If your display has an odd width or height, you may wish to make sure that the origin lies exactly on a pixel:
display.attach(portrayal, "Portrayal Text Goes Here", display.insideDisplay.width / 2,
display.insideDisplay.height / 2, true);
Scaling You might also wish to scale the portrayal. By default the scaling adjusts the portrayal so that its width and height exactly fill the width and height of the window. That is, the following statements are identical:
display.attach(portrayal, "Portrayal Text Goes Here");
display.attach(portrayal, "Portrayal Text Goes Here",
new Rectangle2D.Double(0,0, display.insideDisplay.width, display.insideDisplay.height), true);
Let's say you want to scale the portrayal so that twice its width and twice its height are displayed on-screen. You could instead attach like this:
display.attach(portrayal, "Portrayal Text Goes Here",
new Rectangle2D.Double(0,0, display.insideDisplay.width/2.0, display.insideDisplay.height/2.0), true);
Again, you may wish to scale exactly to a pixel boundary if the height or width of the window is odd:
display.attach(portrayal, "Portrayal Text Goes Here",
new Rectangle2D.Double(0,0, display.insideDisplay.width/2, display.insideDisplay.height/2), true);
Translating and Scaling You can scale and translate the portrayal by providing a Rectangle2D.Double which defines (1) the location of the portrayal origin and (2) the number of pixels that the portrayal's width and height take up by default. This is basically a combination of the above two issues. Let's say you wish to move the origin to the center of the screen and scale out the portrayal by a factor of 10 (so 100 times as much area is displayed in the window). You could write something along the lines of:
display.attach(portrayal, "Portrayal Text Goes Here",
new Rectangle2D.Double(display.insideDisplay.width/2.0, // translated x origin
display.insideDisplay.height/2.0, // translated y origin
display.insideDisplay.width/100.0, // scaled width
display.insideDisplay.height/100.0), // scaled height
true);
...and again, you could change these from doubles to integers to more cleanly adjust to pixel boundaries.
MASON actually has two kinds of inspectors:
)
At present MASON provides plug-in inspectors for streaming a property to a file and for charting the change of a property over time. If you wish to make another inspector, here's what you need to do. Let's say you want to stream a property to standard out.
Last, to tell MASON that it should dynamically load your inspector and plug it in, you need to edit the sim/portrayal/inspector/propertyinspector.classes file and add the classname of your inspector.
The Inspector and PropertyInspector class documentation has further discussion on the issue. Here's a very simple PropertyInspector which writes values out to System.out:
import java.awt.*;
import javax.swing.*;
import sim.display.*;
import sim.engine.*;
import sim.util.*;
import sim.portrayal.inspector.*;
public class StreamOutPropertyInspector extends PropertyInspector
{
double lastTime = Schedule.BEFORE_SIMULATION;
public static String name() { return "Stream to System.out"; }
public static Class[] types() { return null; } // accepts all types
public StreamOutPropertyInspector(Properties properties, int index, Frame parent, GUIState simulation)
{
super(properties,index,parent,simulation);
add(new JLabel("Streaming " + properties.getName(index) + " to System.out"));
validInspector = true; // we always want to go ahead with displaying this inspector
}
public void updateInspector()
{
double time = simulation.state.schedule.time();
if (lastTime < time) // only print if this is a new timestamp
{
lastTime = time;
// print like this... object/property/timestamp: value
System.out.println("" + properties.getObject() + "/" +
properties.getName(index) + "/" +
lastTime + ": " + properties.getValue(index));
}
}
// We always want to make a frame, so we don't override this
// public boolean shouldCreateFrame() { return false; }
// We should flush our stream when we're stopped. So let's wrap the stopper to do this.
public Stoppable reviseStopper(Stoppable stopper)
{
final Stoppable newStopper = super.reviseStopper(stopper);
return new Stoppable()
{
if (newStopper!=null) newStopper.stop(); // wraps the stopper
System.out.flush();
};
}
}
Sometimes it's a pain to squish everything you want to inspect into a single Inspector. MASON provides a class called TabbedInspector which acts as a holder for multiple Inspectors (putting them into a JTabPane). This lets you break your inspector into multiple inspectors to simplify life for you.
It's pretty easy. During your Simple Portrayal's getInspector() method, or while adding a model inspector etc.:
In your GUIState subclass, you'll need to add some instance variables:
org.jfree.data.xy.XYSeries series; // the data series we'll add to
sim.util.media.ChartGenerator chart; // the charting facility
Next, you need to construct the chart in your init method, like so:
public void init(Controller c)
{
super.init(c);
[do your main code here, then...]
chart = new sim.util.media.ChartGenerator();
chart.setTitle("Put the title of your chart here");
chart.setRangeAxisLabel("Put the name of your charted series here");
chart.setDomainAxisLabel("Time");
JFrame frame = chart.createFrame(this);
// perhaps you might move the chart to where you like.
frame.show();
frame.pack();
c.registerFrame(frame);
// the console automatically moves itself to the right of all
// of its registered frames -- you might wish to rearrange the
// location of all the windows, including the console, at this
// point in time....
}
Last, in your GUIState's start() method, you'll need to create a new series and add it to the chart, like this
public void start()
{
super.start();
[do your main code here, then...]
chart.removeAllSeries();
series = new org.jfree.data.xy.XYSeries(
"Put a unique name for this series here so JFreeChart can hash with it",
false);
chart.addSeries(series, null);
scheduleImmediateRepeat(true, new Steppable()
{
public void step(SimState state)
{
// at this stage we're adding data to our chart. We
// need an X value and a Y value. Typically the X
// value is the schedule's timestamp. The Y value
// is whatever data you're extracting from your
// simulation. For purposes of illustration, let's
// extract the number of steps from the schedule and
// run it through a sin wave.
double x = state.schedule.time();
double y = Math.sin(state.schedule.getSteps()) * 10;
// now add the data
series.add(x, y, true);
}
});
}
When you run the simulation, you'll notice that the generator doesn't have all the features of the standard one which pops up with inspectors -- notably you don't have the ability to state how often the chart is updated. These are features of the inspector code, not the chart generator. Your programmatic mechanism is updating the chart every time a value is inserted into it. This is expensive, particularly when you have lots of data points.
If you would like the chart to redraw itself once every second of wall-clock time after values have been sent to it (say), you can write a little timer which fires one second after it's called. Add this to your GUIState subclass:
Thread timer = null;
public void startTimer(final long milliseconds)
{
if (timer == null)
timer= sim.util.Utilities.doLater(milliseconds, new Runnable()
{
public void run()
{
if (chart!=null) chart.update();
timer = null; // reset the timer
}
});
}
Now you just need to tell the series not to update itself automatically when an item is added, and also to turn on the timer (if it's not already on). In the start() method, change the line
series.add(x, y, true);
To...
series.add(x, y, false); // don't update automatically
startTimer(1000); // once a second (1000 milliseconds)