[GSoC 2026] Flent Application API Integration - Week 4 Update
11 views
Skip to first unread message
KEDAR KISAN
Jun 25, 2026, 2:53:04 PM (10 days ago) Jun 25
to ns-developers
Hi everyone,
Here is a brief progress update for Week 4 of the Flent Application API Integration project.
1. CMake & Licensing Cleanup
Updated the nlohmann/json CMake configuration to correctly copy nlohmann/json into the build tree (per the latest tracker patch), rather than only adding third-party/ to the include path. Also added the MIT License file (shared by Tom) to the LICENSES/ directory for the bundled nlohmann/json dependency.
2. Root-Causing the TCP Upload Throughput Discrepancy
Following up on last week's open question about the throughput drop observed on ns-3.40/3.41, I tested the original Flent code against the legacy ns-3.36.1 release and found a flaw in the ProcessRawValues post-processing function: it was not accurately replicating upstream Flent's aggregation logic. Applying the corrected logic to the same ns-3.36.1 simulation produced a throughput of ~41.4 Mbps, the exact value previously observed on ns-3.40. Since the underlying network simulation was unchanged and only the post-processing logic differed, this confirms the discrepancy is a legacy test-bound calibration issue rather than a behavioral regression in ns-3 or in Flent's networking model. Per mentor guidance, focus has shifted from chasing release-to-release alignment toward validating whether 41.4 Mbps is a reasonable throughput baseline for this topology.
3. Testbed Validation
Generated ns-3 Flent plots for TCP Cubic at 10ms and 80ms RTT using flent-example.cc, extended with a --tcpType command-line argument to select congestion control without recompiling. Compared these against reference testbed result plots in a side-by-side comparison document; the plots are broadly similar in shape with some differences, and understanding the exact cause of this gap will be a primary focus next week. you can view the plots using the Link.
You can view the full details and weekly breakdown of the project using the below project links:
- Repository: https://gitlab.com/k.kedar/ns-3-dev (Working Branches: GSoC-2026-flent-application, nlohmann-json)
- ns-3 wiki: https://www.nsnam.org/wiki/GSOC2026Flent
- GSoC Page: https://summerofcode.withgoogle.com/programs/2026/projects/PCXPdFoD
Best Regards,
Kedar
Here is a brief progress update for Week 4 of the Flent Application API Integration project.
1. CMake & Licensing Cleanup
Updated the nlohmann/json CMake configuration to correctly copy nlohmann/json into the build tree (per the latest tracker patch), rather than only adding third-party/ to the include path. Also added the MIT License file (shared by Tom) to the LICENSES/ directory for the bundled nlohmann/json dependency.
2. Root-Causing the TCP Upload Throughput Discrepancy
Following up on last week's open question about the throughput drop observed on ns-3.40/3.41, I tested the original Flent code against the legacy ns-3.36.1 release and found a flaw in the ProcessRawValues post-processing function: it was not accurately replicating upstream Flent's aggregation logic. Applying the corrected logic to the same ns-3.36.1 simulation produced a throughput of ~41.4 Mbps, the exact value previously observed on ns-3.40. Since the underlying network simulation was unchanged and only the post-processing logic differed, this confirms the discrepancy is a legacy test-bound calibration issue rather than a behavioral regression in ns-3 or in Flent's networking model. Per mentor guidance, focus has shifted from chasing release-to-release alignment toward validating whether 41.4 Mbps is a reasonable throughput baseline for this topology.
3. Testbed Validation
Generated ns-3 Flent plots for TCP Cubic at 10ms and 80ms RTT using flent-example.cc, extended with a --tcpType command-line argument to select congestion control without recompiling. Compared these against reference testbed result plots in a side-by-side comparison document; the plots are broadly similar in shape with some differences, and understanding the exact cause of this gap will be a primary focus next week. you can view the plots using the Link.
You can view the full details and weekly breakdown of the project using the below project links:
- Repository: https://gitlab.com/k.kedar/ns-3-dev (Working Branches: GSoC-2026-flent-application, nlohmann-json)
- ns-3 wiki: https://www.nsnam.org/wiki/GSOC2026Flent
- GSoC Page: https://summerofcode.withgoogle.com/programs/2026/projects/PCXPdFoD
Best Regards,
Kedar
Reply all
Reply to author
Forward
0 new messages