| 摘要 |
In the Train algorithm, popular and small oversized objects are defined and placed into separate special cars within the trains. The train and car numbers are arranged to fit into one computer word for simplicity and time efficiency of fetching and comparing. A maximum number of cars numbers allowed in any one train is determined from the counter or register (word) length holding the train/car numbers, and the train and car numbers encompass the entire generation being collected. A threshold of less than the maximum number of cars in one train is determined, and when reached, contiguous special cars are arranged to share a common number, and all the cars in the train are renumbered sequentially. However, cars sharing one car numbers remain as separate data structures, including separate remembered sets. A mechanism for shuffling non-contiguous special cars, so that they are contiguous, is disclosed. Moreover, if needed normal sized cars may be re-numbered and included with other cars in a collection set.
|
