LLM-Network-Study-Data

Per-request network captures (.pcapng) collected by the LLM-Network-Study benchmark harness (benchmark.py and the per-workload test scripts). Each directory holds one capture file per request, named request_<id>_run<n>_<timestamp>.pcapng.

A directory name encodes four dimensions:

<capture-env>_<provider/model>_<workload>[_<dataset/variant>]_results

Dimension legend

Dimension	Values	Meaning
Capture env	ethernet	Wired connection to the cloud API.
	mobile_tethered	Phone-tethered cellular link (phone as modem).
	mobile_untethered	Capture on the mobile device itself over cellular.
	local	Self-hosted model; traffic split into client and server side captures.
Provider	direct (claude_*, openai_gpt_*, gpt_image_1)	Native vendor API (Anthropic / OpenAI).
	openrouter_*	Same model routed through OpenRouter.
	siliconflow_*	SiliconFlow inference provider.
Workload	bfcl_*	Function-calling prompts from the BFCL v4 dataset (see below).
	image_to_text	Vision captioning of DiffusionDB images.
	text_to_image	Image generation from DiffusionDB prompts.
	text_to_text	Plain text generation (local models).
	video_download	Baseline (non-LLM) video download for comparison.
Variant	diffusiondb_{100,405,2k}	Sample size drawn from the DiffusionDB dataset.
	gen1 / gen2	Repeated capture batches of the same config.
	unblocked	Run without rate-limiting/throttling.
	streamed / stream	Streaming run. Inconsistently applied — see Streaming vs non-streaming below; use metadata.csv, not the name.

BFCL workloads come from the Berkeley Function-Calling Leaderboard v4 categories: simple_python (single function call), parallel (multiple calls from one prompt), live_multiple (real-world prompt, pick among many tools), and live_parallel_multiple (real-world prompt, multiple parallel calls among many tools).

Streaming vs non-streaming

Do not infer this from the directory name. Use metadata.csv at the repo root, which carries an explicit streaming column for every one of the 61 capture directories:

import pandas as pd
meta = pd.read_csv("metadata.csv")
streamed = meta[meta.streaming == "true"].directory

streaming	Dirs	Meaning
true	33	Response was delivered incrementally (SSE / chunked token stream).
false	25	Whole response body delivered in one shot.
n/a	3	Non-LLM baseline (video); API streaming does not apply.

Every row also carries streaming_evidence, naming the harness and code path the label came from, so the classification is auditable rather than asserted.

Why the directory name is not reliable

The name encodes streaming inconsistently, in four spellings and positions (_streamed_, _stream_, streamingbfcl, and none at all). Two traps in particular:

• The 16 openrouter_* directories are streaming but carry no stream token. openrouter_testing/benchmark.py sets stream=True unconditionally in run_openrouter_request(). A name-based filter misses all 16.
• ethernet_baseline_video_stream is not a streaming LLM run. It is YouTube video streaming — a workload, not a delivery mode. A grep stream over directory names picks it up wrongly.

Measured columns

measured_median_downlink_gap_ms and measured_frac_gaps_over_5ms are observed from the pcaps (median over ~8 mid-size captures per directory): the inter-arrival gap between downlink payload packets from the server.

These are supporting evidence, not the definition of the label. They separate token streaming cleanly (a slow model such as nemotron_3_super_120b shows ~420 ms gaps; non-streamed bulk delivery shows sub-millisecond gaps), but two confounders make them unusable as a standalone classifier:

• Image generation. A streamed text_to_image run delivers large partial-image chunks at line rate, so it looks like bulk transfer (~0.1 ms gaps) despite streaming.
• Link RTT. Mobile and tethered captures inflate gaps independently of streaming.

The column is blank for 16 directories whose captures have fewer than 10 downlink payload packets — too small to measure, which is itself consistent with a single non-streamed response body.

Datasets

Directory	Capture env	Model / Provider	Workload
ethernet_claude_opus_4_6_bfcl_live_multiple_results	ethernet	Claude Opus 4.6 (Anthropic)	BFCL live_multiple
ethernet_claude_opus_4_6_bfcl_live_parallel_multiple_results	ethernet	Claude Opus 4.6 (Anthropic)	BFCL live_parallel_multiple
ethernet_claude_opus_4_6_bfcl_parallel_results	ethernet	Claude Opus 4.6 (Anthropic)	BFCL parallel
ethernet_claude_opus_4_6_bfcl_simple_python_results	ethernet	Claude Opus 4.6 (Anthropic)	BFCL simple_python
ethernet_gpt_5_4_image_to_text_diffusiondb	ethernet	GPT-5.4 (OpenAI)	Image-to-text (DiffusionDB)
ethernet_gpt_image_1_text_to_image_diffusiondb_100_results	ethernet	gpt-image-1 (OpenAI)	Text-to-image (DiffusionDB, 100)
ethernet_gpt_image_1_text_to_image_diffusiondb_2k_unblocked_results	ethernet	gpt-image-1 (OpenAI)	Text-to-image (DiffusionDB, 2k, unblocked)
ethernet_openai_gpt_5_4_bfcl_live_multiple_results	ethernet	GPT-5.4 (OpenAI direct)	BFCL live_multiple
ethernet_openai_gpt_5_4_bfcl_live_parallel_multiple_results	ethernet	GPT-5.4 (OpenAI direct)	BFCL live_parallel_multiple
ethernet_openai_gpt_5_4_bfcl_parallel_results	ethernet	GPT-5.4 (OpenAI direct)	BFCL parallel
ethernet_openai_gpt_5_4_bfcl_simple_python_results	ethernet	GPT-5.4 (OpenAI direct)	BFCL simple_python
ethernet_openrouter_claude_sonnet_4_6_gen1_bfcl_live_multiple_results	ethernet	Claude Sonnet 4.6 (OpenRouter)	BFCL live_multiple (batch 1)
ethernet_openrouter_claude_sonnet_4_6_gen1_bfcl_live_parallel_multiple_results	ethernet	Claude Sonnet 4.6 (OpenRouter)	BFCL live_parallel_multiple (batch 1)
ethernet_openrouter_claude_sonnet_4_6_gen1_bfcl_parallel_results	ethernet	Claude Sonnet 4.6 (OpenRouter)	BFCL parallel (batch 1)
ethernet_openrouter_claude_sonnet_4_6_gen1_bfcl_simple_python_results	ethernet	Claude Sonnet 4.6 (OpenRouter)	BFCL simple_python (batch 1)
ethernet_openrouter_claude_sonnet_4_6_gen2_bfcl_live_multiple_results	ethernet	Claude Sonnet 4.6 (OpenRouter)	BFCL live_multiple (batch 2)
ethernet_openrouter_claude_sonnet_4_6_gen2_bfcl_live_parallel_multiple_results	ethernet	Claude Sonnet 4.6 (OpenRouter)	BFCL live_parallel_multiple (batch 2)
ethernet_openrouter_claude_sonnet_4_6_gen2_bfcl_parallel_results	ethernet	Claude Sonnet 4.6 (OpenRouter)	BFCL parallel (batch 2)
ethernet_openrouter_claude_sonnet_4_6_gen2_bfcl_simple_python_results	ethernet	Claude Sonnet 4.6 (OpenRouter)	BFCL simple_python (batch 2)
ethernet_openrouter_gemini_3_1_pro_bfcl_live_multiple_results	ethernet	Gemini 3.1 Pro (OpenRouter)	BFCL live_multiple
ethernet_openrouter_gemini_3_1_pro_bfcl_live_parallel_multiple_results	ethernet	Gemini 3.1 Pro (OpenRouter)	BFCL live_parallel_multiple
ethernet_openrouter_gemini_3_1_pro_bfcl_parallel_results	ethernet	Gemini 3.1 Pro (OpenRouter)	BFCL parallel
ethernet_openrouter_gemini_3_1_pro_bfcl_simple_python_results	ethernet	Gemini 3.1 Pro (OpenRouter)	BFCL simple_python
ethernet_openrouter_gpt_5_4_bfcl_live_multiple_results	ethernet	GPT-5.4 (OpenRouter)	BFCL live_multiple
ethernet_openrouter_gpt_5_4_bfcl_live_parallel_multiple_results	ethernet	GPT-5.4 (OpenRouter)	BFCL live_parallel_multiple
ethernet_openrouter_gpt_5_4_bfcl_parallel_results	ethernet	GPT-5.4 (OpenRouter)	BFCL parallel
ethernet_openrouter_gpt_5_4_bfcl_simple_python_results	ethernet	GPT-5.4 (OpenRouter)	BFCL simple_python
ethernet_openrouter_nemotron_3_super_120b_bfcl_live_multiple_results	ethernet	Nemotron 3 Super 120B (OpenRouter)	BFCL live_multiple
ethernet_openrouter_nemotron_3_super_120b_bfcl_parallel_results	ethernet	Nemotron 3 Super 120B (OpenRouter)	BFCL parallel
ethernet_openrouter_nemotron_3_super_120b_bfcl_simple_python_results	ethernet	Nemotron 3 Super 120B (OpenRouter)	BFCL simple_python
ethernet_siliconflow_image_to_text_results	ethernet	SiliconFlow VLM	Image-to-text (DiffusionDB)
ethernet_siliconflow_text_to_image_diffusiondb_405_results	ethernet	SiliconFlow image model	Text-to-image (DiffusionDB, 405)
ethernet_siliconflow_text_to_text_bfcl_parallel_results	ethernet	SiliconFlow LLM	BFCL parallel (text)
ethernet_video_download_results	ethernet	— (baseline)	Video download baseline
local_image_to_text_client_results	local (client side)	Self-hosted VLM	Image-to-text (DiffusionDB)
local_image_to_text_server_results	local (server side)	Self-hosted VLM	Image-to-text (DiffusionDB)
local_text_to_image_diffusiondb_client_results	local (client side)	Self-hosted image model	Text-to-image (DiffusionDB)
local_text_to_image_diffusiondb_server_results	local (server side)	Self-hosted image model	Text-to-image (DiffusionDB)
local_text_to_text_bfcl_parallel_client_results	local (client side)	Self-hosted LLM	BFCL parallel (text)
local_text_to_text_bfcl_parallel_server_results	local (server side)	Self-hosted LLM	BFCL parallel (text)
mobile_tethered_gpt_5_4_image_to_text_results	mobile, tethered	GPT-5.4 (OpenAI)	Image-to-text (DiffusionDB)
mobile_tethered_gpt_image_1_text_to_image_diffusiondb_405_unblocked_results	mobile, tethered	gpt-image-1 (OpenAI)	Text-to-image (DiffusionDB, 405, unblocked)
mobile_tethered_openai_gpt_5_4_bfcl_parallel_results	mobile, tethered	GPT-5.4 (OpenAI)	BFCL parallel
mobile_untethered_gpt_5_4_image_to_text_diffusiondb_100_results	mobile, untethered	GPT-5.4 (OpenAI)	Image-to-text (DiffusionDB, 100)
mobile_untethered_openai_gpt_5_4_bfcl_parallel_results	mobile, untethered	GPT-5.4 (OpenAI)	BFCL parallel

Backup

These directories are mirrored to Box at RPI WAYS LAB/LLM/Result Pcaps/LLM-Network-Study-Data via rclone (rclone copy . "box:RPI WAYS LAB/LLM/Result Pcaps/LLM-Network-Study-Data").