Edge infrastructure traces

These Edge infrastructure traces consist of bandwidth and latency between devices along with the execution time of the microservices of a video processing application on the different devices. These traces were collected between 2022-10-24 and 2022-11-09.

The execution times are related to a set of microservices, including video encoding and framing, along with the training and inference model for road sign classification.The microservices' descriptions are provided in the following research paper:N. Mehran, Z. N. Samani, D. Kimovski, and R. Prodan, "Matching-based Scheduling of Asynchronous Data Processing Workflows on the Computing Continuum," 2022 IEEE International Conference on Cluster Computing (CLUSTER), 2022, pp. 58-70, DOI: 10.1109/CLUSTER51413.2022.00021.

Devices are as follows:- D01 machine has a twelve-core AMD Ryzen Threadripper 2920X processor with 32GB memory;- D02 machine has an eight-core Intel Core(TM) i7-7700 processor with 16GB memory;- Nvidia Jetson Nano machine has a four-core ARM Cortex-A72 processor with 4GB memory;- Raspberry Pi 4 has a four-core ARM Cortex A57 processor with 4GB of memory.

Moreover, the dataset provides the network traces regarding the bandwidth and latency between the Edge devices. We provided the size of the transmitted messages between the devices for the throughput measurements. For the network latency, the information related to the round trip time is calculated by ICMP message request and reply.

The traces include five parts of data as follows:

BW-MessageSize.csv- Timestamp: Date and time in CET- Source device: the throughput from which we are checking; D02 machine- Destination device: the throughput to which we are checking; D01 or D02 machine- Message size (MB): the message size transmitted between the devices- BW (Mbps): the maximum achievable bandwidth*** "-1" means "iperf3: error - unable to connect to server: Connection refused"

JetsonNano-ExecutionTime.csv- Timestamp: Date and time in CET- Device: the Nvidia Jetson Nano single-board computer- Microservice: includes encoding, framing, training, or inference of Dockerized microservices- Execution Time (seconds)

Large-ExecutionTime.csv- Timestamp: Date and time in CET- Device: the highest performance machine (D01 machine) in the C3 testbed- Microservice: includes encoding, framing, training, or inference (containers)- Execution time (seconds)

Latency.csv- Timestamp: Date and time in CET- Source device: the latency from which we are checking; D02 machine- Destination device: the latency to which we were checking; D01 or D02 machine- Minimum latency (ms) of the round-trip times among four transmitted messages- Average latency (ms) of the round-trip times among four transmitted messages- Maximum latency (ms) of the round-trip times among four transmitted messages- Mean standard deviation of the round-trip times among four transmitted messages

RPi4-ExecutionTime.csv- Timestamp: Date and time in CET- Device: Raspberry Pi v4 single-board computer- Microservice: includes encoding, framing, training, or inference (containers)- Execution time (seconds)

For more information regarding the testbed, please refer to the C3 website at https://c3.itec.aau.at/.

Authors:Zahra Najafabadi Samani, Narges Mehran, Dragi Kimovski, Josef Hammer, Radu ProdanInstitute of Information Technology, Alpen-Adria-Universitaet Klagenfurt, Austria

本边缘基础设施跟踪数据集涵盖了设备间的带宽与延迟数据，以及视频处理应用的微服务在不同设备上的执行时长。本数据集采集于2022年10月24日至2022年11月9日期间。

本次采集的执行时长对应一组微服务，包括视频编码、帧处理，以及用于道路标志分类的模型训练与推理流程。相关微服务的详细说明可参阅下述研究论文：N. Mehran, Z. N. Samani, D. Kimovski, and R. Prodan, "Matching-based Scheduling of Asynchronous Data Processing Workflows on the Computing Continuum," 2022 IEEE International Conference on Cluster Computing (CLUSTER), 2022, pp. 58-70, DOI: 10.1109/CLUSTER51413.2022.00021.

本次测试涉及的设备如下：
- D01主机：搭载12核AMD Ryzen Threadripper 2920X处理器，配备32GB内存；
- D02主机：搭载8核英特尔酷睿(TM) i7-7700处理器，配备16GB内存；
- Nvidia Jetson Nano单板计算机：搭载4核ARM Cortex-A72处理器，配备4GB内存；
- 树莓派4（Raspberry Pi 4）：搭载4核ARM Cortex A57处理器，配备4GB内存。

此外，本数据集还提供边缘设备间的网络跟踪数据，包括用于吞吐量测算的设备间传输消息大小。关于网络延迟，其往返时间（RTT）通过ICMP消息请求与应答机制计算得到。

本跟踪数据集包含五部分数据，详情如下：
1. BW-MessageSize.csv
- 时间戳（Timestamp）：采用中欧时间（CET）的日期与时间；
- 源设备（Source device）：待测试吞吐量的发起设备，固定为D02主机；
- 目的设备（Destination device）：待测试吞吐量的接收设备，为D01或D02主机；
- 消息大小（Message size, MB）：设备间传输的消息大小（单位：兆字节）；
- 带宽（BW, Mbps）：可达到的最大带宽。*** "-1"表示"iperf3: 错误 - 无法连接至服务器：连接被拒绝"

2. JetsonNano-ExecutionTime.csv
- 时间戳（Timestamp）：采用中欧时间（CET）的日期与时间；
- 设备（Device）：Nvidia Jetson Nano单板计算机；
- 微服务（Microservice）：包含容器化的视频编码、帧处理、模型训练或推理流程；
- 执行时长（Execution Time, 秒）：单位为秒

3. Large-ExecutionTime.csv
- 时间戳（Timestamp）：采用中欧时间（CET）的日期与时间；
- 设备（Device）：C3测试平台中性能最高的D01主机；
- 微服务（Microservice）：包含容器化的视频编码、帧处理、模型训练或推理流程；
- 执行时长（Execution Time, 秒）：单位为秒

4. Latency.csv
- 时间戳（Timestamp）：采用中欧时间（CET）的日期与时间；
- 源设备（Source device）：待测试延迟的发起设备，固定为D02主机；
- 目的设备（Destination device）：待测试延迟的接收设备，为D01或D02主机；
- 最小延迟（Minimum latency, ms）：4次传输消息的往返时间中的最小值（单位：毫秒）；
- 平均延迟（Average latency, ms）：4次传输消息的往返时间的平均值（单位：毫秒）；
- 最大延迟（Maximum latency, ms）：4次传输消息的往返时间中的最大值（单位：毫秒）；
- 往返时间均值标准差（单位：毫秒）：4次传输消息的往返时间的样本标准差

5. RPi4-ExecutionTime.csv
- 时间戳（Timestamp）：采用中欧时间（CET）的日期与时间；
- 设备（Device）：树莓派4（Raspberry Pi 4）单板计算机；
- 微服务（Microservice）：包含容器化的视频编码、帧处理、模型训练或推理流程；
- 执行时长（Execution Time, 秒）：单位为秒

如需了解更多关于本测试平台的信息，请访问C3官网：https://c3.itec.aau.at/。

作者：Zahra Najafabadi Samani, Narges Mehran, Dragi Kimovski, Josef Hammer, Radu Prodan
奥地利克拉根福阿尔卑斯-亚德里亚大学信息技术学院