A tcp lookup optimization method under high performance computing network

Abstract

The invention discloses a TCP search optimization method under a high performance calculating network; the method comprises the following steps: 1, in TCP conversation processing, preferably calculating TCP conversations to obtain 32-bit short signatures if the TCP conversation order of magnitude is at the million level, carrying out XOR for the front 16-bit and rear 16-bit of the 32-bit short signatures obtained on the previous basis if the TCP conversation order of magnitude is at 100-million level, thus obtaining 16-bit short signatures; 2, respectively using the 32- bit short signatures and the 16-bit short signatures to replace a TCP tetrad identification TCP conversations; 3, building corresponding relations between the front P TCP conversation short signatures and Hash slots one by one; mapping the front P TCP conversation short signatures and Hash slots one by one if arrived TCP conversations exceed P, and assigning the TCP conversation short signatures (starting from P+1) from a TCB pool to a conflict list. The method can reduce the searching taken cache size, and can reduce the Hash conflict probability.

Description

technical field [0001] The invention relates to a TCP search optimization method under a high-performance computing network, belonging to the technical field of communication. Background technique [0002] With the geometric growth of the amount of data in the smart grid, on the one hand, its huge amount of data puts forward huge requirements on the computing power of the equipment and the transmission capacity of the network. In this context, the concept of high-performance computing network came into being. On the other hand, since the smart grid is composed of a large number of power subnetworks, and a large amount of data is transmitted in the network, the data in the smart grid mainly depends on the high-performance computing network for transmission. [0003] At present, the high-performance computing network mainly relies on TCP to implement in the wide area network. However, when the number of TCP sessions reaches a threshold, the performance of the search algorithm ...

AI Technical Summary

Problems solved by technology

And when there are millions of TCP session transmissions in the network, TCB will occupy 260MB 1.3GB of cache space, while the size of the last level cache (LLC) in mainstream commodity processors is usually 10MB, that is to say, TCB will occupy hundreds of thousands of times the size of LLC. order of magnitude, the TCB even needs to occupy TB-level storage space. Therefore, when using the traditional TCP search algorithm to search for TCP sessions, the huge workload will occupy a large cache of the system, and even the phenomenon of insufficient cache may occur, while the system The performance of also degrades rapidly as the number of TCP sessions increases, because the spatial locality of the TCP working set grows proportionally to the number of sessions.

Furthermore, due to the lack of temporal locality of TCB accesses, increasing the cache size brings only limited benefit when a large number of sessions are active, since the lookup time of TCP sessions is mainly determined by the CPU performance of main memory accesses rather than by instruction The execution time of , therefore, the TCP lookup time will be greatly affected by the memory minor access

[0005] In summary, the data structure of the hash table in the traditional TCP lookup algorithm can no longer meet the requirements of finding a large number of TCP sessions in the high-performance computing network, that is to say, the traditional TCP lookup algorithm in the prior art cannot simultaneously handle high-performance network In the case of a large number of TCP sessions, and the sharp deterioration of the performance of the hash table caused by it and the problem of occupying too much cache when looking up

Images