IBM Cloud Code Engine Traces

Data associated with the following publication. Please cite the paper if you use this data in your research.

Nima Nasiri, Nalin Munshi, Simon D Moser, Marius Pirvu, Vijay Sundaresan, Daryl Maier, Thatta Premnath, Norman Böwing, Sathish Gopalakrishnan, and Mohammad Shahrad, "In-Production Characterization of an Open Source Serverless Platform and New Scaling Strategies", 2026 ACM European Conference on Computer Systems (EuroSys '26).

Decompressing the Data

Decompression relies on the 7-zip utility, which you can install in case you do not have it installed already.

• On macOS: brew install p7zip (assuming Homebrew for installation)
• On Ubuntu: sudo apt install p7zip-full p7zip-rar

To decompress the data files, run the decompress.sh script. This will extract all the weekly traffic data files as well as app configuration dataframe into a directory named data. The clone size of this repository is about 6GB, however, after decompression, the size of the data directory will be around 52GB.

bash decompress.sh

Workload Size

• More than 1.9 billion requests
• 1780 applications
• 62 days

Information Included

• Hashed Namespace
• Hashed Application Name
• Daily Application Configurations (see app_configs.pickle)
  • 'NamespaceHash'
  • 'AppHash'
  • 'AppContainerRequestCpu' (in vCPU)
  • 'AppContainerRequestMemory' (in GB)
  • 'AppContainerConcurrency'
  • 'AppMinScale': The minimum number of instances for the application. If set to 0 (default value), the application scales down to zero.
• Weekly Traffic Data (10 weeks, see week_*.pickle)
  • 'NamespaceHash'
  • 'AppHash'
  • 'NumEvents'
  • 'InvocationTimes': A list of size NumEvents containing the invocation timestamps (in seconds)
  • 'AppExecTimes': A list of size NumEvents containing the application execution times (in milliseconds)
  • 'TotalExecTimes': A list of size NumEvents containing the total execution times (in milliseconds)
  • 'PodHash': A list of size NumEvents containing the pod hashes. Each app can have multiple pods, so this column allows to identify which pod served which request. Note that the ordering of requests across the these four lists is consistent.

Using the Data

All data is stored in pickle format and should be loaded using Python Pandas library as DataFrames.

import pandas as pd
df_configs = pd.read_pickle('data/app_configs.pickle')
df_week1 = pd.read_pickle('data/week_1.pickle')
...

Loading all the weekly traffic data files will require a high amount of RAM. Depending on your machine's capabilities, you might want to load part of the data or process the data in chunks.

Using the Sample Scripts

We provide some sample scripts to help you get started with the data. Before running the scripts, run the following command to create a virtual environment and install the required packages.

bash setup.sh

The sample scripts are located in the sample-scripts directory. You would have to run the sample scripts with the sample-scripts as your working directory.