CESNET-USTS23: a benchmark dataset of Unevenly spaced time series from network traffic

This dataset was created to evaluate characteristics of Unevenly sampled time series from network traffic (USTS) for the paper Unevenly Spaced Time Series from Network Traffic.

The file named time_series.tar.gz contains a folder with time series CSV files as raw data of the experiment. In the folder are the following files:

• fts.csv -- contains 2.6 million Flow time series (FTS) created from 259 million IP flows,
• pts.csv -- contains 19 million Packet time series (PTS) created from 110 million network packets,
• sfts.csv -- contains 15 million Single flow time series (SFTS) created from 160 million network packets.

Traffic was captured on the national CESNET2 network from February 2023 to April 2023. All IP addresses in the dataset were anonymized.

The fts.csv has the following format:

• ID_DEPENDENCY -- Identification of a network dependency observed as a Flow time series. (real IP address was anonimized by replacing with a random IP address)
• N_FLOWS -- Number of flows in time series, i.e., number of data points.
• N_PACKETS -- Number of packets in time series, i.e., the sum of metric PACKETS.
• N_BYTES -- Number of bytes in time series, i.e., the sum of metric PACKETS.
• PACKETS -- The array containing the time series metric number of packets in the IP flow.
• BYTES -- The array containing the time series metric number of bytes in the IP flow.
• START_TIMES -- The array containing the time series time axis of the flows starts.
• END_TIMES -- The array containing the time series time axis of the flows ends.

The pts.csv has the following format:

• ID_DEPENDENCY -- Identification of a network dependency observed as a Packet time series. (real IP address was anonymized by replacing with a random IP address)
• BYTES -- The array containing the time series metric payload length of the network packet.
• TIMES -- The array containing the time series time axis of the transmission of network packets.

The sfts.csv has the following format:

• SRC_IP -- Source IP address. (real IP address was anonimized by replacing with a random IP address)
• SRC_PORT -- Source port.
• DST_IP -- Destination IP address (real IP address was anonymized by replacing with a random IP address)
• DST_PORT -- Destination port.
• bytes -- The array containing the time series metric payload length of the network packet.
• time -- The array containing the time series time axis of the transmission of network packets.

The file named characteristics.tar.gz contains a folder with characteristics gained by experiments from time series files. In the folder are the following files:

• fts.characteristics.csv -- Characteristics about Flow time series from the fts.csv.
• pts.characteristics.csv -- Characteristics about Packet time series from the pts.csv.
• sfts.characteristics.csv -- Characteristics about Single flow time series from the sfts.csv.

The fts.characteristics.csv has the following format:

• LENGTH -- Number of data points in the source time series.
• DURATION -- Duration of the source time series.
• H_BYTES -- Hurst exponent of the source time series metric BYTES.
• STATIONARITY_PACKETS -- Stationarity of the source time series metric PACKETS.
• STATIONARITY_BYTES -- Stationarity of the source time series metric BYTES.
• OVERALL_STATIONARITY -- Overal stationarity created by merging STATIONARITY_PACKETS and STATIONARITY_BYTES.

The pts.characteristics.csv and sfts.characteristics.csv have the following format:

• LENGTH -- Number of data points in the source time series.
• DURATION -- Duration of the source time series.
• H -- Hurst exponent of the source time series.
• STATIONARITY -- Stationarity of the source time series.

We provide the samples of all zipped files for a quick lookup: fts.characteristics.sample.csv, fts.sample.csv, pts.characteristics.sample.csv, pts.sample.csv, sfts.characteristics.sample.csv, sfts.sample.csv