Toward Optimal Partial Parallelization for Service Function Chaining
I‐Chieh Lin, Yu-Hsuan Yeh, Kate Ching‐Ju Lin
Abstract
The emergence of Network Function Virtualization (NFV) and Service Function Chaining (SFC) together enable flexible and agile network management and traffic engineering. Due to the sequential execution nature of SFC, the latency would grow linearly with the number of functions. To resolve this issue, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">function parallelization</i> has recently been proposed to enable independent functions to work simultaneously. Existing solutions, however, assume all the function instances are installed in the same physical machine and, thus, can be parallelized with only a little overhead. Nowadays, most of the networks deploy function instances in distributed servers for load balancing, parallelization across different servers would, in fact, introduce a non-negligible cost of duplicating or merging packets. Hence, in this work, we propose PPC ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Partial Parallel Chaining</i> ), which only parallelizes functions if parallelization can indeed reduce the latency after considering function placement and the required additional parallelization cost. To this end, we design two schemes, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">partial parallelism enumeration</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">instance assignment</i> to identify the optimal partial parallelism that minimizes the latency. Our simulation results show that PPC effectively adapts the degree of parallelism and, hence, outperforms both sequential chaining and full parallelism in any general scenario. Overall, the latency reduction can be up to 47.2% and 35.2%, respectively, as compared to sequential chaining and full parallelism.