It models TCP flows interacting with Ethernet Pause (802.3x) in two tandem queues as a Markov process. The TCP’s AIMD is modeled as a MMPP process with $N$ states with monotonically decreasing sending rate. The AIMD is modeled as a move of the state $a\to a-1$ (rate increase) or $a\to 2a$ (rate decrease). The Markove chain is modeled as a four-tuple $(i,j,k,a)$ where $i$ and $j$ are the number of packets in the two queues in tandem (with upper-bound $B_1$ and $B_2$); and $k$ is a binary value denote the “up” and “down” state of the second queue, as whether pause is in effect or not; and $a$ is the state of TCP’s MMPP process.

The paper assumes Poisson arrival of packets and exponential service of queues. Paper also did a simulation with NS2 and OMNeT++. It found that:

- Hop-by-hop flow control improves goodput as RTT increases. Because TCP recovers effectively as RTT is small and HbH control has less room for optimization
- Threshold for sending HbH control shall not be close to buffer size so that you may still have room to catch the burst

## Bibliographic data

```
@inproceedings{
title = "Modeling the interaction of IEEE 802.3x hop-by-hop flow control and TCP end-to-end flow control",
author = "Richa Malhotra and Ronald van Haalen and Michel Mandjes and Rudesindo Núñez-Queija",
howpublished = "NGI",
booktitle = "Proceedings of Next Generation Internet Networks",
year = "2005",
}
```