Minstrel TX Rate Control in User space – GSoC ’22

Introduction

Hi everyone! I’m Prashiddha. I have recently graduated from Jacobs University Bremen with a BSc. Computer Science and minors in Robotics and, Global Economics and Management. For the past year, I have been involved in the research and development of open-source software at SupraCoNeX, primarily focusing on facilitating rate control in user space, which will soon be public.

For GSoC’22, I’ll be working on implementing and testing Minstrel HT, the default WiFi rate control for Linux-based OpenWRT OS, in user space. This first blog post intends to cover details on the necessary background to understand the project and its implementation.

What is WiFi Rate Control?

A typical WiFi network consists of at least a sender and a receiver that communicate through the propagation of radio frequencies within the license-free ISM band. The radio waves carry information in binary as an encoding, and the sender devices can choose from several modulation and transmission parameters such as coding rate, bandwidth, and guard interval. The choice of a transmission scheme between the WiFi devices determines the theoretical network throughput or data rate. A metric called the Modulation Coding Scheme (MCS) Index has been defined to help better understand the WiFi data rates and the RF environment of the network. The MCS index is based on the parameters of the transmission schemes mentioned above.

With newer IEEE 802.11 standards such as IEEE 802.11ax, there are hundreds of available MCS rates for transmission. At first glance, it may seem like maximum data transmission could be easily attained using only those rates which yield the highest theoretical throughput. However, the modulations which achieve high data rates only work best when the link between the WiFi devices is robust. Furthermore, compared to wired-based communication, the wireless communication channel demonstrates higher dynamics and is prone to interference, especially if multiple wireless devices share the medium uncoordinated. As such, the performance of WiFi networks is far from optimal, and there have been significant efforts to develop WiFi rate control algorithms that dynamically adapt transmission data rates in response to the varying wireless channel conditions.

Motivation

In Linux-based OpenWRT WiFi devices, the mac80211 subsystem in the kernel space is responsible for rate control. This includes the implementation of rate control algorithms like the Minstrel HT in the kernel space. The kernel space provides full access to the deviceÔÇÖs hardware and memory. Development of modules is hence subject to the risk of complete system failure due to bugs or failure in particular modules and submodules. Additionally, development in kernel space is restricted to the use of integer value operations. Due to the instability and risk involved in accessing the floating-point unit, floating-point operations are avoided in kernel space. Lastly, capabilities for prototyping and debugging required research and development are highly restricted in this space.

Given the limitations and lack of ease of development in kernel space, the need for a user space rate control algorithm is apparent. With this, my GSoC’22 project will focus on implementing a user space variant of Minstrel HT with experiments designed to compare the performance with its kernel space counterpart.

Deliverables of the project

The end goals of my GSoC ’22 projects are as follows:

  • Software Architecture of the user space Minstrel HT implementation in Python.
  • Proper Documentation and Guide on working with the Minstrel HT package.
  • Ready to run demo script to showcase the potential of user space Minstrel HT.
  • Detailed analysis of WiFi rate control experiments for performance comparison between kernel and user space Minstrel HT.

What’s been already done?

Prior to GSoC ’22, I had already been heavily involved in the implementation of Rateman, short for Rate Manager, which acts as an intermediary to provide necessary information and functionalities for WiFi rate control in user space. The Rateman package is still under development and will soon be released as an open-source OpenWRT package. Furthermore, as a part of my bachelor thesis, I’ve already implemented a working version of the user space Minstrel HT in Python using Rateman.

Since the rate control algorithm in user space needs to be designed such that it can work on multiple access points simultaneously, initially, multiple experiments were conducted to evaluate various methods of parallelizing rate control from the Rateman, namely async task, thread pool, and process pool. The results have indicated async task to be the best scheme for parallelizing rate control from Rateman.

With this, the user space Minstrel HT in Python has been developed to be executed as an async task and the first results seem to indicate towards a comparable performance with its kernel counterpart. However, the first experiments were far from foolproof and require better-designed experiments to yield a concrete result.

What’s next?

The user space Minstrel HT is not complete and still requires implementation of additional features in order to be identical to the kernel Minstrel HT in terms of functionality.

  • Changing the output of user space Minstrel HT to a live printout of the rate statistics table.
  • Adding option to store the output in a CSV file to aid in comparison with the kernel Minstrel HT.
  • Extending user space Minstrel HT with functionalities from the kernel variant such as calculating the number of retransmission, random sample table, and reducing the number of spatial streams.
  • Adding a new estimator called Butterworth Filter which is currently used by the kernel Minstrel HT.
  • Modifying user space Minstrel HT to accommodate for various WiFi rate control experiments if needed.

Concluding Thoughts

With the start of GSoC ’22 coding period, I’ll begin with modifying the output of user space Minstrel HT to the printout of live rate statistics along with the option to save it in a CSV file, identical to the kernel Minstrel HT. This will be followed by extending the user space Minstrel HT with the remaining functionalities from the kernel variant. In the end, the project will mainly focus on experimentations for performance analysis of Minstrel HT in user space.

With this, I’d like to conclude the first blog on the user space Minstrel HT GSoC ’22 project with freifunk. Please feel free to reach out and connect with me. ­čÖé

Thanks for reading!

One thought on “Minstrel TX Rate Control in User space – GSoC ’22

Leave a Reply

Your email address will not be published. Required fields are marked *