Efficient hierarchical hash tree for OpenFlow packet classification with fast updates on GPUs

Lin, Yi-Sheng; Shih, Wen-Chi; Chang, Yeim-Kuan

doi:10.1016/j.jpdc.2022.04.018

Cited by 2 publications

(1 citation statement)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Because the reduced dependence of data and control makes them more appropriate for the parallelism on multi-core and many-core systems. Many researchers have attempted to use GPUs to solve computing intensive problems in related domains [13][14][15][16][17][18][19][20][21][22][23][24][25]. Similarly, each filter is independent while executing conflict detection, and the field content matching does not require complex computations.…”

Section: Introductionmentioning

confidence: 99%

An Efficient Parallel Algorithm for Detecting Packet Filter Conflicts

2022

View full text Add to dashboard Cite

Advanced network services, such as firewalls, policy-based routing, and virtual private networks, must rely on routers to classify packets into different flows based on packet headers and predefined filter tables. When multiple filters are overlapped, conflicts may occur, leading to ambiguity in the packet classification. Conflict detection ensures the correctness of packet classification and has received considerable attention in recent years. However, most conflict-detection algorithms are implemented on a conventional central processing unit (CPU). Compared with a CPU, a graphics processing unit (GPU) exhibits higher computing power with parallel computing, hence accelerates the execution speed of conflict detection. In this study, we employed a GPU to develop two efficient algorithms for parallel conflict detection: the general parallel conflict-detection algorithm (the GPCDA) and the enhanced parallel conflict-detection algorithm (the EPCDA). In the GPCDA, we demonstrate how to perform conflict detection through parallel execution on GPU cores. While in the EPCDA, we analyze the critical procedure in conflict detection as to reduce the number of matches required for each filter. In addition, the EPCDA adopts a workload balance method to enable load balancing of GPU execution threads, thereby significantly improving performance. The simulation results show that with the 100 K filter database, the GPCDA and the EPCDA execute conflict detection 2.8 to 13.9 and 9.4 to 33.7 times faster, respectively, than the CPU-based algorithm.

show abstract