onvm

This directory contains some explorative experimental results related to the high-speed OpenNetVM platform on a COTS server.

Introduction

This directory contains our explorative experiments of the state-of-the-art OpenNetVM (ONVM) platform on a commodity off-the-shelf server. Our experiments focus on the performance diagnoses of ONVM. Please refer to the main website for further information. The general architecture of ONVM is illustrated below:

The experiments include 5 different Service Function Chaining (SFC) configurations, namely:

1. Linear
2. DAG-1
3. DAG-2
4. Tree
5. Bridge

We conduct three types of experiments:

1. Load stimulus: Perturbing the input traffic with different input rates and patterns.

2. Resource stimulus: Perturbing the allocated resource to individual VNFs. Currently, we assign collocated parasite processes to each VNF to perturb its CPU share.

3. The combination of the previous two experiments.

Please refer to each SFC directory for data and further details.

How to interpret the data

CSV files for each experiment:

Files omnipresent in all experiments:

• tx_stats: Statistics for the input traffic. The input traffic was generated and recorded using the MoonGen traffic generator. The generated synthetic packets follow either pure 128 B or the Internet IMIX distribution (i.e., 64B: 512B: 1024B = 7:4:1). Please refer to the tx_stats.csv file in each directory to check the packet size distribution. Normally, the packet rate of IMIX should not surpass 4 Mpps (i.e., million packets per second). A sample tx_stats record is illustrated below:

Time	Device	Direction	PacketRate	Mbit	MbitWithFraming	TotalPackets	TotalBytes
1686141358	Device: id=0	TX	0.33343282345416	943.37429415621	996.72354590888	339888	120205980

• rx_stats: End-to-end throughput statistics are also measured and recorded using MoonGen. This file follows the same format as tx_stats. A sample rx_stats record is illustrated below:

Time	Device	Direction	PacketRate	Mbit	MbitWithFraming	TotalPackets	TotalBytes
1686141399	Device: id=1	RX	0.18560265302008	525.69664884533	555.39307332854	15963959	5653132404

• latency: Statistics for the end-to-end latency. This is also measured using MoonGen by injecting PTP packets every second into the tx traffic. The latency is measured in nanoseconds.

Sequence no.	Latency (ns)
188	3765443

• nf: Statistics of individual VNFs measured by Perf. The interesting features include CPU frequency, l1-dcache-misses, and LLC cache references, and so on.

Files specific to the ONVM platform

• nf_out: This file is output by ONVM to measure the statistics of the individual VNFs on a configured SFC. Some interesting statistics include the TX/RX rates, RX/TX counts, TX/RX drop rates, TX/RX drops. The column names are listed as follows:

nfs[i].tag, nfs[i].instance_id, nfs[i].service_id, nfs[i].thread_info.core, rx_pps, tx_pps, rx, tx, act_out, act_tonf, act_drop, nfs[i].thread_info.parent, state, rte_atomic16_read(&nfs[i].thread_info.children_cnt), rx_drop_rate, tx_drop_rate, rx_drop, tx_drop, act_next, act_buffer, act_returned

Future (or ongoing) directions:

Please refer to our paper if you find this work interesting:

Qiong Liu, Tianzhu Zhang, Leonardo Linguaglossa. "Non-invasive performance prediction of high-speed softwarized network services with limited knowledge," IEEE International Conference on Computer Communications (INFOCOM), 2024.

Collaboration Acknowledgement

This project involves a collaborative effort between Telecom Paris and Nokia Bell Labs. Our partnership reflects a shared commitment to advancing open-source software development, data science research, and applied AI/ML in networking.

We extend our gratitude to all contributors and supporters who have made this project possible. If you would like more details, questions, or suggestions about our project, please feel free to contact us at [email protected] or [email protected].