Abstract: Concurrent cuckoo hashing is performed on a hash table that includes a number of locations; each may hold a value. A plurality of processes may concurrently execute on the table; each process includes a sequence of operations, which are divided into a number of phases. Each phase corresponds to one operation in the sequence. An overflow buffer is provided for each location in the table. Each overflow buffer may hold a value displaced from its corresponding location in the table. A plurality of sequences of operations is concurrently executed. Each phase in a sequence executes by acquiring one or two locks on two locations in the table; a lock acts on a location and its overflow buffer. An operation of a phase is then executed. If, on conclusion of the phase execution, any overflow buffer holds a value, the execution is repeated until all overflow buffers are empty.

Abstract: A set of structures and techniques are described herein whereby an exemplary concurrent shared object, namely a shared skip list, can be implemented in a lock-free manner. Indeed, we have developed a number of interesting variants of a lock-free shared skip-list, including variants that may be employed to provide a lock-free shared dictionary. In some variants, a key-value dictionary is implemented.

Abstract: A multiple-location read, single-location write operation is implemented using transient blocking synchronization support. The multiple-location read, single-location write operation involves first acquiring transient ownership of a memory location to be modified and then acquiring transient ownership of at least one other memory location, the contents of which are read and used to modify the memory location first acquired.

Abstract: A concurrent FIFO queue is implemented as an “optimistic” doubly-linked list. Nodes of the optimistic doubly-linked list are allocated dynamically and links between the nodes are updated optimistically, i.e., assuming that threads concurrently accessing the FIFO queue will not interfere with each other, using a simple store operation. Concurrent, linearizable, and non-blocking enqueue and dequeue operations on the two ends of the doubly-linked list proceed independently, i.e., disjointly. These operations require only one successful single-word synchronization operation on the tail pointer and the head pointer of the doubly-linked list. If a bad ordering of operations on the optimistic FIFO queue by concurrently executing threads creates inconsistencies in the links between the nodes of the doubly-linked list, a fix-up process is invoked to correct the inconsistencies.

Abstract: A deque of a local process in a memory work-stealing implementation may use one or more data structures to perform work. If the local process attempts to add a new value to its deque's data structure when the data structure is full (i.e., an overflow condition occurs), the contents of the data structure are copied to a larger allocated data structure (e.g., an array of greater size than an original array). The entries in the original, smaller-sized data structure are copied to exact positions in the now-active, larger-sized data structure. By this technique, the local process is thus provided with space to add the new value.

Abstract: A computer system implementing transient blocking synchronization allows a memory location leased by a first process to be read-accessible to another process. In other words, more than one thread may have read-only type leases on a given memory location at a given time. Such “shared” leases expire when respective lease periods of the shared leases elapse.

Abstract: A computer system uses transient blocking synchronization for performing operations on shared memory. When performing operations on more than one memory location, the computer system obtains transient exclusive access to a first memory location. The computer system then obtains transient exclusive access to a second memory location, where the transient exclusive access to the second memory location does not expire prior to an expiration of the transient exclusive access to the first memory location or until explicitly unleased.

Abstract: A method for inserting an object into a concurrent set including obtaining a key associated with the object, traversing the concurrent set using a first thread containing the key, identifying a first insertion point while traversing the concurrent set, where the first insertion point is before a current node and after a predecessor node, obtaining a first lock for the predecessor node after identifying the first insertion point, validating the predecessor node and the current node after obtaining the lock, inserting a new node into the concurrent set after validating, where the new node is associated with the object, and releasing the first lock after inserting the new node.

Abstract: A method for coordinating shared access to data objects comprises applying modification operations to a data object from a first thread of a plurality of threads on behalf of all the other threads during a session in which the first thread owns a lock on the data object. Each modification operation corresponds to a respective entry recorded in a log associated with the data object by a respective thread. The method may further comprise predicting, for a second thread, a result of a particular operation requested by the second thread on the data object. The result may be predicted using log entries corresponding to modification operations that have not yet been applied to the data object. In addition, the method includes performing one or more other operations in a non-blocking manner from the second thread during the session, where at least one other operation is dependent on the predicted result.

Abstract: A method comprises associating a plurality of locks with a data object accessed concurrently by a plurality of threads, where each lock corresponds to a respective partition of the object. The method includes using a first non-blocking transaction (such as a Hardware Transactional-Memory (HTM) transaction) to attempt to complete a programmer-specified transaction. The first non-blocking transaction may access one or more of the locks but may not actually acquire any of the locks. In response to an indication that the first non-blocking transaction failed to complete, the method may include acquiring a set of locks in another non-blocking transaction, where the set of locks corresponds to a set of partitions expected to be accessed in the programmer-specified transaction. If the set of locks is acquired, the method may include performing the memory access operations of the programmer-specified transaction, and releasing the set of locks.

Abstract: The design of nonblocking linked data structures using single-location synchronization primitives such as compare-and-swap (CAS) is a complex affair that often requires severe restrictions on the way pointers are used. One way to address this problem is to provide stronger synchronization operations, for example, ones that atomically modify one memory location while simultaneously verifying the contents of others. We provide a simple and highly efficient nonblocking implementation of such an operation: an atomic k-word-compare single-swap operation (KCSS). Our implementation is obstruction-free. As a result, it is highly efficient in the uncontended case and relies on contention management mechanisms in the contended cases. It allows linked data structure manipulation without the complexity and restrictions of other solutions.

Abstract: The present disclosure describes a unique way for each of multiple processes to operate in parallel and use the same shared data without causing corruption to the shared data. For example, during a commit phase, a corresponding transaction can attempt to increment a globally accessible version information variable and store a current value of the globally accessible version information variable for updating version information associated with modified data regardless of whether an associated attempt by the corresponding transaction to modify the globally accessible version information variable was successful. As an alternative mode, a corresponding transaction can merely read and store a current value of the globally accessible version information variable without attempting to update the globally accessible version information variable before such use.

Abstract: A “garbage collector” employed to reclaim memory dynamically allocated to data objects employs multiple execution threads to perform a parallel-execution operation and its garbage-collection cycle. A thread executes tasks that it selects from lists whose entries represent tasks dynamically identified during other tasks' performance. When a thread fails to find a task in one of these lists, it sets to an inactivity-indicating value a field associated with it in a global status word. It also determines whether any field associated with any of the other threads indicates activity. If not, the thread concludes that the parallel-execution operation has been completed. Otherwise, it returns to searching for further tasks to perform.

Abstract: In response to source code that represents instructions for dynamically allocating memory to objects, a compiler/interpreter produces instructions that implement a garbage collector. The garbage collector operates in garbage-collection cycles, which include parallel-execution operations such as locating reachable objects. Each thread maintains a respective task queue onto which it pushes identifiers of objects thus found and from which it pops those identifiers in order to begin the tasks of locating the further objects to which objects specified by the thus-popped identifiers refer. A thread's access to its respective task queue ordinarily occurs on a last-in, first-out basis, but the access mode switches to a first-in, first-out basis if the number of task-queue entries exceeds a predetermined threshold.

Abstract: A linked-list-based concurrent shared object implementation has been developed that provides non-blocking and linearizable access to the concurrent shared object. In an application of the underlying techniques to a deque, the linked-list-based algorithm allows non-blocking completion of access operations without restricting concurrency in accessing the deque's two ends. The new implementation is based at least in part on a new technique for splitting a pop operation into two steps, marking that a node is about to be deleted, and then deleting it. Once marked, the node logically deleted, and the actual deletion from the list can be deferred. In one realization, actual deletion is performed as part of a next push or pop operation performed at the corresponding end of the deque.

Abstract: A garbage collector employs a plurality of task queues for a parallel-execution operation in a garbage-collection cycle. Each task queue is associated with a different ordered pair of the threads that perform the parallel-execution operation in parallel. One of the threads, referred to as that task queue's “enqueuer” thread, is the only one that can “push” onto that queue an identifier of a dynamically identified task. The other thread, referred to as that task queue's “dequeuer,” is the only one that can “pop” tasks from that task queue for execution. Since, for each task queue, there is only one thread that can “push” task identifiers on to it and only one thread that can “pop” task identifiers from it, the garbage collector can share dynamically identified tasks optimally among its threads without suffering the cost imposed by making combinations of otherwise separate machine instructions atomic.

Abstract: A sequentially performed implementation of a compound compare-and-swap (nCAS) operation has been developed. In one implementation, a double compare-and-swap (DCAS) operation does not result in a fault, interrupt, or trap in the situation where memory address A2 is invalid and the contents of memory address A1 are unequal to C1. In some realizations, memory locations addressed by a sequentially performed nCAS or DCAS instruction are reserved (e.g., locked) in a predefined order in accordance with a fixed total order of memory locations. In this way, deadlock between concurrently executed instances of sequentially performed nCAS instructions can be avoided. Other realizations defer responsibility for deadlock avoidance to the programmer.

Abstract: A novel linked-list-based concurrent shared object implementation has been developed that provides non-blocking and linearizable access to the concurrent shared object. In an application of the underlying techniques to a deque, non-blocking completion of access operations is achieved without restricting concurrency in accessing the deque's two ends. In various realizations in accordance with the present invention, the set of values that may be pushed onto a shared object is not constrained by use of distinguishing values. In addition, an explicit reclamation embodiment facilitates use in environments or applications where automatic reclamation of storage is unavailable or impractical.

Abstract: A multiprocessor, multi-program, stop-the-world garbage collection program is described. The system initially over partitions the root sources, and then iteratively employs static and dynamic work balancing. Garbage collection threads compete dynamically for the initial partitions. Work stealing double-ended queues, where contention is reduced, are described to provide dynamic load balancing among the threads. Contention is resolved by using atomic instructions. The heap is broken into a young and an old generation where parallel semi-space copying is used to collect the young generation and parallel mark-compacting the old generation. Speed and efficiency of collection is enhanced by use of card tables and linking objects, and overflow conditions are efficiently handled by linking using class pointers. A garbage collection termination employs a global status word.

Abstract: A system is disclosed for providing, from an input data word comprising a plurality of input data units having an input arrangement and a mask word comprising a plurality of mask bits each associated with one of the data units, an output data word in which the data units are arranged according to the mask bits. The system includes a bit balancer module and a plurality of rearrangement modules. The bit balancer module is configured to divide the input data units comprising the input data word into a plurality of data word portions, each data unit being assigned to one of the data word portions based on a pattern of mask bits of the mask word relative to the mask bit associated with the respective data unit. Each rearrangement module is configured to provide, from one of the data word portions and associated mask bits, an output data word portion in which the data units are arranged according to the mask bits.