Lock-Free Search Data Structures: Throughput Modeling with Poisson Processes

This paper considers the modeling and the analysis of the performance of lock-free concurrent search data structures. Our analysis considers such lock-free data structures that are utilized through a sequence of operations which are generated with a memoryless and stationary access pattern. Our main contribution is a new way of analyzing lock-free concurrent search data structures: our execution model matches with the behavior that we observe in practice and achieves good throughput predictions. Search data structures are formed of basic blocks, usually referred to as nodes, which can be accessed by two kinds of events, characterized by their latencies; (i) CAS events originated as a result of modifications of the search data structure (ii) Read events that occur during traversals. An operation triggers a set of events, and the running time of an operation is computed as the sum of the latencies of these events. We identify the factors that impact the latency of such events on a multi-core shared memory system. The main challenge (though not the only one) is that the latency of each event mainly depends on the state of the caches at the time when it is triggered, and the state of caches is changing due to events that are triggered by the operations of any thread in the system. Accordingly, the latency of an event is determined by the ordering of the events on the timeline. Search data structures are usually designed to accommodate a large number of nodes, which makes the occurrence of an event on a given node rare at any given time. In this context, we model the events on each node as Poisson processes from which we can extract the frequency and probabilistic ordering of events that are used to estimate the expected latency of an operation, and in turn the throughput. We have validated our analysis on several fundamental lock-free search data structures such as linked lists, hash tables, skip lists and binary trees