Performance Evaluation of MPTCP Incast Based on Queuing Network

Multipath TCP (MPTCP) is an important protocol, which is used to solve the incremental communication mode in wireless communication networks (WCNs) and data center networks (DCNs) realizing the balance of data transmission and increase the aggregation bandwidth. However, With the wide application of multi host hosts, in the FatTree topology, a performance bottleneck exists that causes throughput collapse in the Incast communication pattern leading to the collapse and seriously affects the network energy. In this paper, based on the queuing network and the Markov property of the MPTCP scheduling process, we propose a multilevel cooperative MPTCP Incast performance evaluation model whose levels are called M/M/N/
${\mathrm {m}}^{\mathtt {I}} {\rightarrow }\text{M}$
/M/L/
${\mathrm {m}}^{\mathtt {II}} {\rightarrow }\text{M}$
/M/K/
${{\mathrm {m}}^{\mathtt{III}}}$
to study the delay performance in MPTCP Incast throughput collapse. The M/M/N/
${\mathrm {m}}^{\mathtt {I}} {\rightarrow }\text{M}$
/M/L/
${\mathrm {m}}^{\mathtt {II}} {\rightarrow }\text{M}$
/M/K/
${{\mathrm {m}}^{\mathtt {III}}}$
levels respectively describe the three levels of data packet arrival, which include the bottleneck links and transmission hotspot Top-of-Rack (ToR) switches in the edge layer, the bottleneck links in the aggregation layer, and the bottleneck links in the transmission layer. This model can calculate the average delay of end-to-end data transmission, which can be used to better analyze the delays involved in the MPTCP Incast throughput collapse. Using NS3 software, we verified the accuracy of the proposed models. The experimental results show that the estimated delays of the proposed model are close to the actual measurement delays and that the edge layer delay proportion is higher. These results help to verify the MPTCP Incast performance bottleneck in the edge layer and provide a theoretical basis for further MPTCP Incast research.