FRAPpuccino: Fault-detection through Runtime Analysis of Provenance

• DOI: 10.48550/arxiv.1711.11487
• Publication Date: 2017-01-01
• Authors: Xueyuan Han; Thomas Pasquier; Tanvi Ranjan; Mark Goldstein; Margo Seltzer
• Keywords: Sliding window protocol

We present FRAPpuccino (or FRAP), a provenance-based fault detection mechanism for Platform as Service (PaaS) users, who run many instances of an application on large cluster machines. FRAP models, records, and analyzes the behavior its impact system directed acyclic provenance graph. It assumes that most behave normally uses their to construct model legitimate behavior. Given behavior, dynamic sliding window algorithm compare new instance's execution model. Any instance does not conform is identified anomaly. prototype experimental results showing it can accurately detect anomalies.