Introduction
Hello everyone!
I’m Prashiddha, a former GSoC contributor with Freifunk in 2022 during which I extended the Py-Minstrel-HT rate control to make it further comparable with its kernel counterpart, allowing for better experimentation between the rate controls in user space and kernel space. If you would like to know more about WiFi rate control and my previous project, please feel free to start with the introduction blog from 2022.
For GSoC ’23, I’ll be working on the research and development of a resource allocation algorithm that can select the optimum transmission rates in conjunction with the optimum power level. The joint power and rate control algorithm is intended to work on OpenWRT routers, capable of making resource allocation decisions for each station connected with it.
Overview of Joint Power and Rate Control
A rate control algorithm, such as Minstrel-HT, determines the best transmission rates that are promising in providing the maximum throughput for the given link condition. These algorithms usually assign a high static power level which could potentially cause interference, especially in a highly dense network. It is already evident that, for a transmission rate, even though a higher transmission power implies a higher signal-to-noise ratio (SNR), it doesnβt necessarily mean higher throughput. Hence, it could be best to use the lowest transmission-power level that is still capable of providing the optimum throughput. As such, this could allow for better management of interference along with an increase in spatial reuse.
The graph presented in a dissertation from Prof. Thomas HΓΌhn shows the relation between the power level and measured throughput where the throughput stops increasing after a certain power level.
WiFi Resource Allocation in Userspace
As part of the SupraCoNeX research, the development of Open-source Resource Control API (ORCA) for OpenWrt access points, has enabled WiFi resource allocation from the user space. The API exposes relevant information from the mac80211 kernel subsystem, such as supported Modulation Coding Scheme (MCS) rates and packet counts (ACKs), that could be required by resource allocation algorithms to make decisions. Previously, the ORCA API could be used to only set the MCS rates for wireless transmission, however, with the recent extension, it allows the MCS rates to be set in conjunction with power levels. Consequently, it is now possible to develop a joint rate and power controller in user space.
In order to further facilitate resource allocation, a Python-based package called “Rateman” has also been developed which utilizes the minstrel-rcd to concurrently operate on multiple access points and parse the exposed kernel information from the API. The package is implemented such that the resource allocation algorithms can be executed through it while also providing them with the parsed kernel information for decision-making.
Extending Py-Minstrel-HT with power control
Since a rate control algorithm in user space already exists, namely “py-minstrel-ht”, I plan on extending the user space Minstrel-HT algorithm with an additional capability for transmit power tuning, also making it convenient to test the effects of power tuning on a rate control algorithm. The main idea behind the joint controller is to let Minstrel-HT decide the set of the best rates while a power tuning module tweaks the power levels to an optimal value. With the addition of power control, the user space Minstrel-HT can be executed with different power settings to achieve various goals. For instance, three different power modes could be realized: fixed power, maximum throughput, and power ceiling.
The fixed power and power ceiling modes are straightforward to understand and implement. The fixed power mode, as the name suggests, sets the power level of all the transmission rates to the specified value. Similarly, the power ceiling mode can be used to specify the maximum power level that can be used for wireless transmission. However, the maximum throughput mode is a bit complicated as the wireless channel is highly dynamic in nature and the controller needs to accurately assess the quality of the link in real-time. Hence, the implementation needs to be well thought-out for every part of the user space Minstrel-HT so as to not hamper the optimal throughput. As the addition of power control adds another depth to the sampling parameter, the set of possible sampling candidates will grow tremendously. However, as Minstrel-HT already probes with a frequency of 50 Hz or 20 ms, sampling too much can greatly degrade the overall performance of the link.
Deliverables
- Extension of py-minstrel-ht with a power controller with complete documentation and execution guide.
- Ready to run demo scripts to showcase the potential of the joint rate and power control.
- Evaluation of the joint controller by comparing it at different modes and with different rate controls.
What’s Next?
At the beginning of the GSoC ’23 coding period, I’ll start by modifying the Rateman package such that the rate statistics dictionary is properly structured to relay successes and attempts statistics per power level per rate. Consequently, I will modify the Py-Minstrel-HT to accommodate the change in the rate statistics structure. This would allow algorithms to better assess the performance of an MCS rate at different power levels. Furthermore, I will extend the rate setting and probing functions from Py-Minstrel-HT to enable power annotation for a desired rate.
Initially, the power ceiling and fixed power modes will be implemented in order to make testing out the power tuning easier. For this, the Py-Minstrel-HT will also be extended to parse the power setting specified by the user in the rc_opts dictionary. If possible, the following questions could also be investigated before the implementation of the max throughput mode:
- Is the power setting completely static with kernel Minstrel-HT? Does the driver play any role in independent power adjustment?
- In general, is the throughput vs tx-power graph strictly non-decreasing? Is it possible that an MCS rate works at power level πππ1 but not at πππ2 where πππ1 < πππ2?
- In a Minstrel-HT rate group, let π 1 and π 2 be two rates where π 2 is a higher rate than π 1. If π 2 works at ππ1, does it imply that π 2 also works at ππ1?
With this, Iβd like to conclude the first blog on the joint power and rate controller in user space. Thanks for reading! Please feel free to reach out and connect with me π
One thought on “GSoC ’23: Joint Power and Rate Control in Userspace for Freifunk OpenWrt Mesh & Access Networks”