ServerlessEdgeTaskOffloadingDataset

Abstract

## Dataset: Serverless Edge Task Offloading Workload A synthetic workload dataset of **5,000 tasks** for evaluating task-offloadingdecisions in serverless edge computing. Each row represents one task generatedby an IoT device, together with the network and device conditions at thatmoment. The dataset was generated with a fixed random seed (42), so it can bereproduced exactly using the included generator script. > Note: this is a synthetic dataset. The file name `azure_style_dataset.csv`> refers to its serverless timing model; it is **not** the Azure Functions> production trace. ### Composition | Task type | Records | Share ||-----------|---------|-------|| Image | 2,033 | 40% || Video | 1,497 | 30% || NLP | 1,470 | 30% | ### Task characteristics | Parameter | Image | Video | NLP ||-----------|-------|-------|-----|| Data size (median) | 7.4 MB | 24.5 MB | 1.0 MB || CPU cycles (median) | 3.0 × 10⁹ | 3.6 × 10⁹ | 1.8 × 10⁹ || Memory (mean) | 512 MB | 1024 MB | 256 MB || Base deadline | 3.0 s | 8.0 s | 1.5 s || Cold-start probability | 0.38 | 0.45 | 0.25 || Mean cold-start time | 0.892 s | 1.25 s | 0.45 s | - Data size and CPU cycles follow log-normal distributions; memory follows a normal distribution (minimum 64 MB).- Each task's deadline is its base deadline scaled by its CPU demand relative to the task-type median, clipped to 0.5–2 times the base. ### Columns | Column | Description ||--------|-------------|| `task_id`, `task_type` | Task identifier and type (image, video, nlp) || `data_size` | Input data size (MB) || `cpu_cycles` | Required CPU cycles || `memory_required` | Memory requirement (MB) || `deadline` | Task deadline (s) || `priority`, `gpu_need` | Task priority and GPU requirement (0–1) || `iot_cpu_power`, `iot_battery` | IoT device CPU frequency (Hz) and battery level (%) || `efaas_core_power`, `efaas_load` | Edge FaaS core frequency (Hz) and load || `bandwidth`, `network_delay` | Uplink bandwidth (Mbps) and network delay (s) || `channel_gain`, `noise` | Wireless channel parameters || `local_execution_time`, `transmission_time` | Pre-computed local and transfer times (s) || `is_cold_start`, `cold_start_time` | Cold-start flag and delay (s) || `exec_time_serverless`, `serverless_latency` | Serverless execution time and total latency (s) || `edge_execution_time`, `total_offload_time` | Edge execution and total offloading time (s) || `local_energy`, `offload_energy` | Monetary cost of local and offloaded execution ($) || `optimal_action` | Reference label: 0 = local, 1 = partial, 2 = full offload | ### Usage