Tarp-data

Tarp experimental data

We collected the experimental data through a series of online experiments and saved it. You can directly refer to the open source project tarp on Github[1] to run the experiment without collecting the data again. If you want to know the details of the experimental data and the collection process, please contact me by email.

Directory structure

Introduction

We implemented the SVM-RL adopted by Firm [2] and CNN-RL adopted by Sinan [3] as control comparison approaches .

CNN-RL: uses CNN to extract features ofstatic topology and use RL to provision resources for key microservice

SVM-RL: uses SVM to detect key microservice and use RL to provision resources for key microservice

Ours: the method proposed in this paper

dgl_graph: Graph dataset in dgl format, used for gnn training, for Experiments I and II (effectiveness on key microservice detection and transferability)

gnn: The resulting data from Experiments I and II (effectiveness on key microservice detection and transferability)

load: Considering the complexity of the real environment of microservice applications, we use workloads by referring to the existing typical open source workload provided by Alibaba [4].

reward # Reward data for the reinforcement learning process (Experiments III )

Directory struct

│ Readme.md # readme file│├─CNNRL # Raw data for CNN-RL│ avg-160.csv│ avg.csv│ p95-160.csv│ Readme.md│ resource.csv│ tp-80.csv│├─dgl_graph # Graph dataset in dgl format│ │ 40-test.g # graph data for trainticket│ ││ ├─ms # graph data for media service│ │ 1.g│ │ 11.g│ │ 2.g│ │ 3.g│ │ 9.g│ │ no.g│ ││ └─ss # graph data for social network│ 1.g│ 2.g│ 3.g│ 4.g│ 5.g│ 6.g│ no.g│├─gnn # The resulting data from Experiments I and II│ acc.csv # accuracy source data │ loss.csv # losssource data │├─load # simulation data of workloads│ load-b.csv│ load-b.png│ load.png│ load1.csv│ load1.png│ load2.png│ load3.csv│├─ours # Raw data for our method│ avg.csv│ resource.csv│ rewards-50.csv│ slo.csv│├─reward # Reward data for the reinforcement learning process│ reward.csv│ reward.png│└─SVMRL # Raw data for SVM-RL avg-1.csv p95-1.csv resource-1.csv rewards-1.csv slo-1.csv tp-160.csv

References

[1] Tarp project. https://github.com/Zenghao-CQ/tarp.

[2] H. Qiu, S. S. Banerjee, S. Jha, Z. T. Kalbarczyk, and R. K. Iyer. FIRM: An Intelligent Fine-grained Resource Management Framework for SLO-Oriented Microservices. In OSDI.

[3] Y. Zhang, W. Hua, Z. Zhou, G. E. Suh, and C. Delimitrou. Sinan: ML-based and QoS-aware resource management for cloud microservices. In ASPLOS, pages 167–181. ACM, April 2021.

[4] Alibaba Cluster Trace Program. https://github.com/alibaba/clusterdata.