System and method for improving processing performance of Bloom filter by using xeon Phi coprocessor

Abstract

The invention discloses a system and a method for improving processing performances of a Bloom filter through utilizing a Xeon Phi coprocessor, wherein the system comprises a performance sampling module, a task dispatch module, a communication module and a task processing module. The performance sampling module is used for obtaining processing capabilities at a host end and a coprocessor end, and deciding a proportion of tasks assigned to the two ends; the task dispatch module controls the overall task dispatch between the host end and the coprocessor end; the communication module manages the communication between the host end and the coprocessor end; and the task processing module takes charge of inquiring and calculating the tasks. The system assigns each task to a corresponding thread; each thread processes the task with a private subvector and converts a serial Hash operation of a single task into a parallel Hash operation of multiple tasks. The method is executed by a linear manner following the procedures of task dispatch, communication and task processing, with the synchronization overhead in the task processing stage being eliminated by the efficient task dispatch, and with the throughput of the task processing being improved at the same time by the efficient parallel mechanism.

Description

technical field [0001] The invention belongs to the technical field of multi-core / many-core performance optimization, and more specifically relates to a system and method for improving the processing performance of a Bloom filter by using a Xeon Phi coprocessor. Background technique [0002] Bloom Filter (Bloom Filter) is a random data structure with high space efficiency, proposed by Burton Bloom in 1970, which uses a bit vector to represent a set, and maps a single element to the set using k mutually independent Hash functions in the bit vector. When inserting an element, set the positions of the k mappings on the bit vector to 1; when searching for an element, judge whether the k positions are all 1. This data structure can judge whether the element to be queried exists in the collection in a very short time. Compared with other data structures such as search trees, Hash tables, etc., the search efficiency and space efficiency decrease continuously with the increase of ...

AI Technical Summary

Problems solved by technology

Due to the increase in the number of cores, a more complex architecture and memory system are adopted. On the one hand, it is not efficient when processing applications with many synchronous communication operations.

At present, most of the commonly used parallel processing methods of Bloom filters use synchronous operations to ensure the correctness of the results, so although many-core coprocessors can support the simultaneous execution of more threads than multi-core processors, once a single thread After entering the critical section of the Bloom filter, more threads will be in a waiting state, and cannot fully perform; on the other hand, due to the independence of each Hash function in the Bloom filter, for each physical As far as threads are concerned, the characteristics of memory access are random, and the locality of the Cache cannot be effectively utilized, and there is a consistency problem with the Cache of other physical threads. In order to maintain the consistency of the cache, more overhead is brought, resulting in Overall memory access efficiency is low

Images