Split-Ordered Lists: Lock-Free Extensible Hash Tables (a summary)

The hash table is a common data structure for applications requiring constant time insert, delete and find of certain items. The basic idea is to use an array of buckets, where each bucket stores a list of items. A hash function is used to determine which bucket a certain item belongs to. Finding or deleting a specific item amounts to a linear search in a bucket list pointed to by the hash function, which can be considered a constant time operation if the list is small. When the number of items grows, constant time operations cannot be guaranteed, so the hash table needs to be extended. A common way of doing this, is to increase the size of the bucket array and redistribute the items among the new buckets. In the context of concurrent programming, the operations find, delete and resize requires special care since several parallel threads may be accessing the data structure simultaneously. Previous work on concurrent hash tables have used lock-based schemes to allow safe operation on individual buckets, or sets of buckets. These implementations have shown reasonable performance, but they do not scale well for increasing number of concurrent accesses. Also, in order to maintain the average constant time complexity, the lock-based schemes tend to be very complex, especially for the resize operation.