Prof. Dr. Frank Bellosa

Daniel Habicht 

Yussuf Khalil

Abstract
Many applications with strong data persistence requirements rely on fsync() to actively persist data and ensure the consistency of file contents on disk. However, fsync() is associated with many problems regarding performance, reliability, and the risk of data loss or corruption. In order to improve the design of interfaces like fsync(), more knowledge about how applications use them in practice is required. Building on top of well-established tracing mechanisms for the Linux kernel, we present a novel approach to examining the synchronous writeback behavior of applications and implement it for the ext4 file system. We examine the synchronous writeback behavior of different consumer applications, and present our results using a visualization technique that has not previously been applied
to data about synchronous writeback. Based on our results, we provide several insights, including that most synchronization requests cause little to no writeback and that writeback latency is not always proportional to the amount of writeback. Our tooling is based on a modular design and can be extended in various ways to include new data sources and answer different high-level questions about the writeback behavior of applications. By building on our approach, future research can provide a better understanding of application writeback behavior in order to improve the design of both kernel interfaces and applications.
BibTex:

@bachelorthesis{besau25AnalyzingApplicationWriteback,

   author = {David Besau},

   title = {Analyzing Application Writeback Behavior},

   type = {Bachelor Thesis},

   year = 2025,

   month = apr # "~3",

   school = {Operating Systems Group, Karlsruhe Institute of Technology (KIT), Germany}

  }