Muyuan Shen's GSoC2023 Proposal
输入“/”快速插入
Muyuan Shen's GSoC2023 Proposal
飞书用户1792023年6月2日创建
About me
About the project
Title
ns3-ai enhancements
Size
Large (350 hours)
Summary
The proposed project aims to enhance the ns3-ai module, which provides interfaces between ns-3 and Python-based ML frameworks using shared memory, with a focus on performance optimization and expanding the range of supported data structures.
To achieve this, the project will introduce APIs for additional data structures like vector and string in shared memory IPC to reduce the interaction between C++ and Python. Additionally, the project will provide examples demonstrating how to implement ML algorithms within ns-3 using C++ and open-source frameworks such as TensorFlow and PyTorch. The project will also improve the current examples and documentation and integrate new examples, such as LTE handover.
Overall, the project aims to expand and accelerate the capabilities of the ns3-ai module, enabling users to simulate and analyze large-scale networks with greater efficiency and flexibility.
User-visible changes
1.
Improved examples and documentation
Introduction
One issue with the current version of ns3-ai is that it was originally developed based on ns-3.30, and as a result, not all examples can be built with newer releases of ns-3 using the CMake build system. Additionally, the documentation for the module is somewhat abstract, leaving curious developers wanting more information about implementation details.
To address these issues, the proposed project aims to revise existing examples one by one, fixing any bugs and adding further specifications to the documentation. The project will also introduce additional examples, such as LTE-handover, to enhance the diversity of the module and provide users with a better understanding of how to use the ns3-ai module effectively.
Examples & documentation to be enhanced
•
q-tcp
•
interference-pattern
Examples & documentation to be added
•
LTE-handover
•
Wifi rate adaptation with DRL
•
Wifi Multi-BSS optimization with DRL
1.
C++ vector and string supported in shared memory
Introduction
Currently, the ns3-ai module allows for sharing C-style structures between the C++ and Python sides. However, support for STL containers and data structures, such as vector, is not yet available. This limitation can be especially problematic when tackling research problems such as network resource allocation, as the lack of vector support forces the user to decompose large vectors into multiple structures and set them separately in shared memory. This approach involves numerous memory reading and writing operations, which can be costly.
In addition, the absence of string support in C++ can also pose challenges for users who need to pass log information to the Python side. Without string support, debugging becomes more difficult. To address these issues, my proposed project aims to provide the following APIs:
•
vector support: The project will provide APIs to enable vector support in shared memory IPC, allowing users to pass large amounts of data more efficiently.
•
string support: The project will provide APIs to enable string support in shared memory IPC, making it easier for users to pass log information and facilitate debugging.
Overall, my project aims to enhance the ns3-ai module by enabling support for vector and string data types, which will improve the efficiency and convenience of shared memory IPC.
Vector sharing API (for reference)
•
C++ side
On the C++ side, users should be able to define vectors of different data types, such as double or int, and specify the vector length to pass data to shared memory. The maximum length of the vector should also be defined to reserve the necessary amount of memory in the shared memory segment.
struct Env
{
int len1;
std::vector<double> name1;
}Packed;
•
Python side
class Env(Structure):
pack = 1
fields = [
('len1', c_int),
('name1', vector),
]
String sharing API (for reference)
•
C++ side
On the C++ side, users must define the string. Since the length can be accessed by method length(), there is no need to specify the size of memory. However, the maximum length should be defined for reservation purposes in case of memory overflow in the shared memory segment. I also provide a tag field to facilitate the classification of different types of information (e.g., debug level).
struct Log