Your browser does not support JavaScript!

Post-graduate theses

Current Record: 44 of 824

Back to Results Previous page
Next page
Add to Basket
[Add to Basket]
Identifier 000452537
Title Kernel level support for transparent use of huge-pages in memory mapped I/O
Alternative Title Υποστήριξη επιπέδου πυρήνα για την διαφανή χρήση μεγάλων σελίδων σε είσοδο/έξοδο χαρτογραφημένης μνήμης
Author Μαλλιωτάκης, Ιωάννης Π.
Thesis advisor Μπίλας, Άγγελος
Reviewer Μαγκούτης, Κωνσταντίνος
Πρατικάκης, Πολύβιος
Abstract Memory-mapped I/O (mmio) allows applications to transparently access data in storage devices via the page-fault mechanism, using processor load/store in- structions. mmio has the potential to (a) eliminate modifications to applications for handling and processing large datasets by merely extending their heap over fast storage devices and (b) provide attractive abstractions for an application to control its I/O path over a unified data representation. Despite these advantages, mmio has significant limitations that make it less attractive. In this thesis, we first discuss the current limitations of mmio. Then, we design xmap, an alterna- tive mmio implementation for the Linux kernel, which addresses these limitations. The main contributions of xmap are support for transparent huge pages over block- based storage and asynchronous promotions. To our knowledge, xmap is the first system that provides this support for the Linux kernel. We evaluate xmap with a variety of graph processing workloads using Ligra, an in-memory graph processing framework, by transparently extending its heap over storage with no code modifi- cations. Our results show that when processing graph datasets 6-8× larger than the available system DRAM, xmap outperforms Linux mmap by up to 3.5×, re- duces total page faults by up to 265×, decreases CPU system time by up to 90%, and increases CPU user time by up to 250%
Language English
Subject Memory management
Operating systems
Storage systems
Διαχείριση μνήμης
Λειτουργικά συστήματα
Συστήματα αποθηκεύσης
Issue date 2022-12-02
Collection   School/Department--School of Sciences and Engineering--Department of Computer Science--Post-graduate theses
  Type of Work--Post-graduate theses
Permanent Link https://elocus.lib.uoc.gr//dlib/b/5/3/metadata-dlib-1669914454-332913-5279.tkl Bookmark and Share
Views 472

Digital Documents
No preview available

Download document
View document
Views : 10