The analytical framework for optimizing TCP retransmission algorithm based on adjustment of its EWMA parameters
In this paper we derive mathematical description of TCP (Transmission Control Protocol) retransmission probability based on Jacobson’s smoothing algorithm that belongs to EWMA (Exponentially Weighted Moving Average) category. This description is parametrized on the probability density function (pdf)...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Polish Academy of Sciences
2024-03-01
|
| Series: | International Journal of Electronics and Telecommunications |
| Subjects: | |
| Online Access: | https://journals.pan.pl/Content/130717/31-4414-Janowski-sk-new.pdf |
| Summary: | In this paper we derive mathematical description of TCP (Transmission Control Protocol) retransmission probability based on Jacobson’s smoothing algorithm that belongs to EWMA (Exponentially Weighted Moving Average) category. This description is parametrized on the probability density function (pdf) of RTT (Round Trip Time) samples and α, β – two primary parameters of Jacobson’s algorithm. Although it is not a close form expression, it is formulated as an effective algorithm that let us to explicitly calculate the values of RTO (Retransmission Time Out) probability as a function of α, β and the pdf of RTT samples. We achieve the effectiveness of this approach by applying smart discretization of the state space and replacement of continuous functions with discrete approximate equivalents. In this way, we mitigate the cardinality of discrete distributions we deal with that results in linear (n+m) instead of multiplicative (n⋅m) growth of computational complexity. We provide the evaluation of RTO probability for a wide set of α, β parameter values and differently shaped Normal and Laplace pdfs the RTT samples are taken from. The obtained numerical results let us to draw some conclusions regarding the choice of optimal values of α, β parameters as well as the impact of pdf the RTT samples are taken from. |
|---|---|
| ISSN: | 2081-8491 2300-1933 |