一种高性能计算网络下的TCP查找哈希算法

A High Performance Computing Network TCP Optimization Algorithm Based on Hash Table

在计算机与通信网络紧密结合的时代,智能电网中的数据处理需要依靠计算机集群来完成,数据传输主要依靠高性能计算网络完成,而高性能计算网络在广域网中主要依赖于TCP协议来实现。由于一般的基于哈希表的TCP查找算法的性能会在TCP会话过多的情况下急剧恶化,且会导致计算机查找TCP会话时产生缓存占用过多的情况,因此提出了一种优化后用来支持高性能计算网络中计算机查找大量TCP会话的高效的TCP查找算法。该算法主要对计算机通过哈希函数生成TCP会话的哈希值的方法以及哈希表的数据结构和映射方式两方面进行优化,并实现了一种适合现代计算机体系的数据结构。为了验证该算法的性能,在Intel多核处理器上进行了并行化堆栈。实验表明,该算法减少了大量TCP会话情况下计算机的TCP会话查找时间和占用的计算机缓存大小,并能在并行平台上同时处理百万级个会话。

In the era of the close connection between computer and the communication network, the data processing in the smart grid needs to be done by computer clusters,the data transmission is mainly done by the high performance computing network,and the high perform- ance computing network mainly depends on TCP in the wide area network. As the performance of the typical hash based TCP lookup al- gorithm is drastically deteriorated in the case of too many TCP sessions, it will contribute to the situation where too much cache of the computers is occupied by TCP sessions. Therefore, we present a high-efficient TCP lookup algorithm that alms at supporting large num- ber of sessions in high performance computing network. It mainly optimizes the TCP session in beth the way of generating the hash value and the data structure and mapping mode of the hash table that well fits the modem computer architectures well. In order to verify the per- formance of the algorithm, we parallelize the stack on the Intel multi core processors. Experiments show that the algorithm reduces the TCP session lookup time and the cache size of a computer in the situation of a large number of TCP sessions, and can handle millions of conversations on a parallel platform at the same time.