Entries are enqueued using the put() method, and dequeued using the take() method. The get() method returns the smallest entry in the queue, but without removing it from the queue. The toArray() methods can be used to examine the contents of the queue.
This class operates like a 'bag' or multiset collection. This simply means that an entry can be added into the queue even if it already exists in the queue. If a 'set' behavior is desired, one can subclass CalendarQueue and override the put() method.
The queue works as follows. Entries are conceptually stored in an infinite set of virtual bins (or buckets). The instance of CQComparator is consulted to determine which virtual bin should be used for an entry (by calling its getVirtualBinNumber() method). Each virtual bin has a width, which can be altered by calling the setBinWidth() method of the CQComparator. Within each virtual bin, entries are sorted.
Having an infinite number of bins, however, is not practical. Thus, the virtual bins are mapped into physical bins (or buckets) by a modulo operation. If there are n physical bins, then virtual bin i maps into physical bin i mod n.
This is analogous to a calendar showing 12 months. Here, n = 12. An event that happens in January of any year is placed in the first month (bin) of this calendar. Its virtual bin number might be year*12 + month. Its physical bin number is just month.
The performance of a calendar queue is very sensitive to the number of bins, the width of the bins, and the relationship of these quantities to the entries that are observed. Thus, this implementation may frequently change the number of bins. When it does change the number of bins, it changes them by a specifiable bin count factor. This defaults to 2, but can be specified as a constructor argument. Suppose the bin count factor is binCountFactor and the current number of buckets is n (by default, this starts at 2, but can be specified by a constructor argument, minNumBuckets). The number of bins will be multiplied by binCountFactor if the queue size exceeds n * binCountFactor. The number of bins will be divided by binCountFactor if the queue size falls below n/binCountFactor. Thus, the queue attempts to keep the number of bins close to the size of the queue. Each time it changes the number of bins, it uses recently dequeued entries to calculate a reasonable bin width (actually, it defers to the associated CQComparator for this calculation).
For efficiency, this implementation constrains minNumBuckets and binCountFactor to be powers of two. If something other than a power is two is given, the next power of two larger than the specified number is used. This has the effect of ensuring that the number of bins is always a power of two, and hence the modulo operation used to map virtual bin numbers onto actual bin numbers is a simple masking operation.
Changing the number of bins is a relatively expensive operation, so it may be worthwhile to increase binCountFactor to reduce the frequency of change operations. Working counter to this, however, is that the queue is most efficient when there is on average one event per bin. Thus, the queue becomes less efficient if change operations are less frequent. Change operations can be entirely disabled by calling setAdaptive() with argument false.
This implementation is not synchronized, so if multiple threads depend on it, the caller must be. This implementation is based on:
| |
| |
| |
| |
| |
| |