 ICACT20220040 Slide.01
[Big Slide]
|
Chrome  Click!! |
 |
Good afternoon everyone. I am Yao-Chung Chang from National Taitung University Taitung, Taiwan. It’s my great pleasure to be here online to share this paper with you.
The paper title is “Low-Latency Computation Offloading based on 5G Edge Computing Systems”. |
| ICACT20220040 Slide.02
[Big Slide]
|
Chrome Click!! |
 |
This study proposed the LSOPG algorithm to improve the quality of GaaS.
In this study, Data Collection mainly collected UE (User Equipment), MEC, and cloud resource information.
According to the Data Collection report, Services Simulation Model can simulate the service transmission and execution latency.
According to simulation results, Offload Model can avoid service congestion caused by queuing and waiting and get the best offloading policy.
|
| ICACT20220040 Slide.03
[Big Slide]
|
Chrome Click!! |
 |
5G and edge computing reduce the time for data to and from the cloud, significantly improve the service quality of the virtual experience, and improve the problem of delay affecting the experience.
Therefore, this study proposed the low-latency computation offloading based on a 5G edge computing system, dynamically configuring the network service's execution environment, improving the overall operating efficiency of the network service.
|
| ICACT20220040 Slide.04
[Big Slide]
|
Chrome Click!! |
 |
This figure describes the Low-Latency Computation Offloading based on the 5G Edge Computing System.
First, the UE establishes a connection with the GaaS server,
and the MEC obtains resource information of the UE, the MEC of the collaboration space, and the cloud.
Then the UE sends the action to the GaaS server for logical processing.
After processing, a set of T of computation tasks will be generated and returned to the MEC for simulation and find a low latency offloading policy.
The computation task will be sent to the location designated by the offloading decision for execution.
The processed data will return to the MEC and encoded into ahttps://www.icact.org/images/b_write.gif real-time video stream to the UE.
Finally, the UE decodes and displays the real-time video stream.
UE and gNB are connected by 5G NR, and Xn converges the CU (MEC) in the collaborative space. |
| ICACT20220040 Slide.05
[Big Slide]
|
Chrome Click!! |
 |
Next is the part of the system architecture, which is divided into 3 Models in total.
The first Model is Data Collection, which is responsible for obtaining UE, MEC and cloud information.
The second Model is the Services Simulation Model, which is responsible for simulating service transmission delays and computing delays.
Finally, there is the Offload Model, which is responsible for finding the best offloading scheme and minimizing the overall Latency. |
| ICACT20220040 Slide.06
[Big Slide]
|
Chrome Click!! |
 |
The Data Collection Model is divided into three Modules.
The first is the UE Resource Information Collector Module, which mainly collects UE resource information, including CPU, GPU, RAM, and computing workload.
Resource information is used to simulate the processing time required to perform services in the UE.
|
| ICACT20220040 Slide.07
[Big Slide]
|
Chrome Click!! |
 |
The MEC Resource Information Collector Module mainly sends resource information requests to the MEC Hosting Infrastructure Management System, and collects all MEC resource information in real time, including MEC-ID, CPU, GPU, RAM, workload, and resource utilization status.
Resource messages are used to simulate the processing time required to execute services in MECs and the resource usage of each MEC.
|
| ICACT20220040 Slide.08
[Big Slide]
|
Chrome Click!! |
 |
The Cloud Resource Information Collector Module mainly sends resource information requests to Cloud Management Platform, and then requests computing resource information from Cloud Resource Management to collect cloud resource information in real time, including CPU, GPU, RAM, computing workload, and location.
|
| ICACT20220040 Slide.09
[Big Slide]
|
Chrome Click!! |
 |
Transmission Latency Simulator is used to simulate the transmission latency.
Execution Latency Simulator is used to simulate the service execution latency of the service in UE, MEC, and cloud.
|
| ICACT20220040 Slide.10
[Big Slide]
|
Chrome Click!! |
 |
The last Model is the Offload Model for finding the best offloading strategy.
When the computing task is offloaded from the MEC to other servers, the network will incur communication costs. , the following three scenarios were considered in this study.
The first scenario is that, assuming the lowest delay in MEC A processing, the UE will send a resource request to MEC A through 5G NR.
After the task is processed, MEC A will send the data back to the UE through 5G NR.
|
| ICACT20220040 Slide.11
[Big Slide]
|
Chrome Click!! |
 |
The second scenario is that, assuming that the processing in MEC B is the lowest latency, the UE will send a resource request to MEC A through 5G NR, and then MEC A will transmit the task to MEC B through Xn for processing.
Xn transmits data back to MEC A, and then transmits it back to UE through 5G NR
|
| ICACT20220040 Slide.12
[Big Slide]
|
Chrome Click!! |
 |
The third scenario is, assuming that the processing in the cloud is the lowest latency, the UE will send a resource request to MEC A through 5G NR, and then MEC A will send the task to the cloud through the fiber for processing.
After the processing is completed, the cloud will send the data back through the fiber Pass it to MEC A, and then pass it to UE through 5G NR
|
| ICACT20220040 Slide.13
[Big Slide]
|
Chrome Click!! |
 |
How to know which one has the smallest delay in the above three scenarios, so we will convert the problem into an objective function, and find the best shunting strategy through the objective function
The part of the first line is the objective function:
Condition 1 and Condition 2 are the setting of the upper limit of the delay
Condition 3 and Condition 4 are MEC resource usage status
Condition 5 is to ensure that each computing task j is executed only once
Condition 6 is to ensure that the offloading policy value is binary
|
| ICACT20220040 Slide.14
[Big Slide]
|
Chrome Click!! |
 |
Finally, there is the pseudo code of the Low Latency Services Offloading Policy based on Greedy (LSOPG) algorithm proposed by this research.
|
| ICACT20220040 Slide.15
[Big Slide]
|
Chrome Click!! |
 |
This experimental environment has a cloud server, 1 main MEC, 3 regional cooperative MECs and 20 UEs to perform 1080p, 60fps stream video services.
The experiment is divided into three groups: One experimental group is the LSOPG algorithm proposed in this study and two control groups are the greedy algorithm and the preset algorithm using the NS-3 simulation software.
The relevant parameter settings are referred to the following paper
|
| ICACT20220040 Slide.16
[Big Slide]
|
Chrome Click!! |
 |
This is the result of the experiment, in terms of average latency, in the context of 20 UEs, the proposed LSOPG algorithm can improve the E2E Latency by about 1.92% compared to the greedy algorithm.
|
| ICACT20220040 Slide.17
[Big Slide]
|
Chrome Click!! |
 |
From the perspective of Packet Loss Rate, LSOPG algorithm has 22.25% and 26.81% improvement compared with greedy algorithm and preset algorithm
Further, if the user wants a better streaming service experience, we need more computing and processing resources and reduce the packet loss rate to achieve the ideal experience service.
|
| ICACT20220040 Slide.18
[Big Slide]
|
Chrome Click!! |
 |
Next, the second experiment was to increase the number of users and analyze the results of the method
|
| ICACT20220040 Slide.19
[Big Slide]
|
Chrome Click!! |
 |
In terms of average latency, increasing the number of users increases the E2E latency slightly, but it is within the required range.
|
| ICACT20220040 Slide.20
[Big Slide]
|
Chrome Click!! |
 |
From the perspective of Packet Loss Rate, increasing the number of users, Packet Loss Rate increased from 41.39% to 43.67%.
Moreover, the LSOPG algorithm still has 35.32% and 35.51% improvement compared to the greedy algorithm and the default algorithm
|
| ICACT20220040 Slide.21
[Big Slide]
|
Chrome Click!! |
 |
Conclusions
Conclusions
Proposed a low-latency computation offloading based on 5G edge computing system for Game as a Service (GaaS).
This study proposes a low latency services offloading policy based on greedy (LSOPG) algorithm.
E2E latency and packet loss rate in the 1080P@60fps scenario (20UEs) can improve 1.92% and 22.25% ; in the 1080P@60fps scenario (50UEs) can improve 2.41% and 35.32%.
|
| ICACT20220040 Slide.22
[Big Slide]
|
Chrome Click!! |
 |
Thank You for Listening!
If you have any questions about this study, please e-mail me. |
