Question about network slicing

[HanchuZhou]

Very impressive work! I would like to ask if it is possible to implement my RL algorithm on network slicing problem? In the network slicing, the action will be the allocation [a1,a2,a3, ... ,ak] of resource to K slices. Thank you for your help.

[GeminiLight]

Thank you for your appreciation. Could you provide some references on network slicing that clearly define the problem? After thoroughly understanding them, I will attempt to unify this task into the Virne Library.

[HanchuZhou]

Since I am not majored in communication, I may not able to give a very precise and professional description. I am trying to conduct an experiment on a RL algorithm for network slicing. Consider there are M users in K slices, the algorithm aims to allocation limited resource to K slices to fullfil maximum latency and minimum throughput requirements of users. I expect the simulator could simulate the system dynamic with varying user's requirement that the result (latency, throughput) is produced after my action (allocation) is taken. I believe the dynamic does not need to be extremely complex. Hope this description is appropriate. I am still looking for a simple simulator for communication system and I believe your simulator with many built-in algorithm will help me for comparison. Thank you again for your time.

For the setting, I believe this paper explains a simple setting of 5G slice: Priority-Based Dynamic Resource Allocation Scheme in Network Slicing https://ieeexplore.ieee.org/abstract/document/9333944

[ychga]

Very impressive work! I would like to ask if it is possible to implement my RL algorithm on network slicing problem? In the network slicing, the action will be the allocation [a1,a2,a3, ... ,ak] of resource to K slices. Thank you for your help. Definitely you can

[GeminiLight]

Recently, I read a paper from JSAC-2024 titled 'SAST-VNE: A Flexible Framework for Network Slicing in 6G Integrated Satellite-Terrestrial Networks.' This paper presents a formulation of graph mapping that resembles the Virtual Network Embedding (VNE) problem. Therefore, I would like to explore the essential differences between these two relevant problems. If you have any insightful answers, please let me know. Thanks!

[AlexisSchOrange]

From what I get, Network Slicing can be seen as VNE with additional, more applied constraints (latency, virtual functions, etc.). But I don't think Network Slicing is a "well-defined" problem, its definition can change a bit from paper to paper. I think the term is used by more applied teams, whereas poeple that are more leaning on the graphs aspects will call it VNE. As we coined it with my teachers, VNE is the "core problem" behind Slicing. (i'm not so knowledgable in communications tho so if anyone has some insights)

[GeminiLight]

Thanks. I agree with most of your insights. The latency feature is still under development due to its complex modeling, and we expect it to be available in November. By then, our framework will be more extensible for network slicing.

[AlexisSchOrange]

How to intend to model it ? I think usually it's done with each virtual edge having a latency threshold, and the latency of the path routing that virtual edge (which is the sum of the latencies of the substrate edges) must be lower. I'm working in exact solving, and it is a linear constraint, perhaps in the (many!) other methods you implemented it is more complicated

[AlexisSchOrange]

Also, there is this thing where the slices are separated into three types (URLLC, eMBB, mMTC), it could be amazing if you had some way of generating different type of slices (tho the reliability and the bandwidth are also considered, which can make it quite more hard !)

[zhiweiyi11]

I think the most difference between VNE and network slicing is that the service request model, the former is about a virtual graph, while the latter is about a Service Function Chain for each request. Though they can be similar from the Graph model, but the key of slicing mapping is about each SFC mapping since they are assumed independent logically. This paper gives good illustration about relationship between the slicing and VNE in NFV-enabled networks, may you can ref it. "Resource Allocation in NFV: A Comprehensive Survey" https://ieeexplore.ieee.org/abstract/document/7534741

[AlexisSchOrange]

That's a good reference, thx; To clarify how I see it, VNE is a graph problem, and Network Slicing is a telco technology. Slicing might be modeled through VNE, or it might be through Service Function Chaining (SFC, the problem that you are describing); but the technology is not working yet, so it's quite hard to know how to exactly model it. A previous PhD in my team modeled it this way I think: "A slice is a superposition of several intertwined SFC", thus they are only partly independant. But I will try to dig a bit in the following weeks to have a clearer mind on this.

[GeminiLight]

That's a good reference, thx; To clarify how I see it, VNE is a graph problem, and Network Slicing is a telco technology. Slicing might be modeled through VNE, or it might be through Service Function Chaining (SFC, the problem that you are describing); but the technology is not working yet, so it's quite hard to know how to exactly model it. A previous PhD in my team modeled it this way I think: "A slice is a superposition of several intertwined SFC", thus they are only partly independant. But I will try to dig a bit in the following weeks to have a clearer mind on this.

Looking forward to hearing more of your insights!