Container-Traffic-Prediction

This is a basic container traffic predicition using ARIMA model.

Our TNSM paper will be published in IEEE soon.

Z. Wu, Y. Deng, H. Feng, Y. Zhou, G. Min and Z. Zhang. Blender: A Container Placement Strategy by Leveraging Zipf-like Distribution within Containerized Data Centers. IEEE Transactions on Network and Service Management. IEEE, Accepted.

1. Main idea

The main idea of predicting the traffic matrix is shown on below:

It depicts a scenario where two historical traffic matrixes (i.e., $T_1$ and $T_2$ ) are used to predict the next traffic matrix $T_3$.

Each block in the traffic matrix indicates the total traffic volume between two containers.

For example, $T_3(1,2)$ can be predicted by using $T_1(1,2)$ and $T_2(1,2)$ as training dataset with the ARIMA model.

By analogy, other values in can be predicted with the same approach.

2. Data source

We collect this traffic trace from a public microservice (robot-shop), where the communication architecture as follow:

3. Schema

Field	Comment
timestamp	time stamp
ip_src	source IP of container
ip_dst	destination IP of container
traffic	traffic volume gerenated by the IP pair

4. Getting strated

• Run TrafficMatrixPrediction.ipynb with jupyter notebook

5. Data discussion

The file traffic-composite_placements.csv also represented the traffic between containers follows a Zipf-like distribution, which the disscusion is included in our paper (Section III).

The distribution as like as:

6. Checking result

• See pdf saved in ./images/