File size: 2,213 Bytes
c7d759d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
# 🛰️ Reveal: Curated Telemetry Dataset for Machine Learning Infrastructure Profiling and Anomaly Detection

**Authors:** Ziji Chen, Steven W. D. Chien, Peng Qian, Noa Zilberman  
**Institution:** University of Oxford  
**Paper:** [Detecting Anomalies in Systems for AI Using Hardware Telemetry (arXiv, 2025)](https://arxiv.org/abs/submit/6934461)  
**License:** [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)  
**Version:** 1.0  

---

## 📘 Overview

**Reveal** is a large-scale, curated telemetry dataset for studying performance profiling, anomaly detection, and infrastructure optimization in modern machine learning (ML) systems.  
It captures **hardware-level signals** from both GPU-equipped clusters while running **over 30 popular ML workloads** across NLP and computer vision domains.

This dataset was collected using the **Reveal** framework, a hardware-centric profiling and unsupervised anomaly detection system introduced in our paper.  
Reveal observes **CPU, GPU, memory, network, and storage** metrics without requiring access to user workloads, making the dataset ideal for operator, side anomaly detection and system performance analysis.

---

## 🧠 Motivation

Modern ML systems are tightly coupled across hardware and software layers, yet operators often lack visibility into workloads due to virtualization and containerization.  
Reveal bridges this gap by providing a **hardware-only telemetry view**, enabling anomaly detection and performance diagnosis **without application instrumentation**.

---

## 🧩 Dataset Description

| Category | Description |
|-----------|-------------|
| **Sampling rate** | 100 ms per metric |
| **Metric types** | ~150 raw metric types per host |
| **Subsystems covered** | CPU, GPU, Memory, Network, Disk |
| **Time-series channels** | ~700 per host |
| **Workloads** | 30+ ML applications including BERT, BART, ResNet, ViT, VGG, DeepSeek, LLaMA, Mistral |
| **Datasets used** | GLUE/SST2, WikiSQL, PASCAL VOC, CIFAR, MNIST |
| **Systems** | Dual-host GPU HPC cluster |
| **Telemetry tools** | `perf`, `procfs`, `nvidia-smi`, Linux utilities |

Each record corresponds to a **time-series window** of low-level system metrics.

---