DOCSLIB.ORG

Umap : Enabling Application-Driven Optimizations for Page Management Ivy B

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Umap : Enabling Application-Driven Optimizations for Page Management Ivy B UMap : Enabling Application-driven Optimizations for Page Management Ivy B. Peng Marty McFadden Eric Green [email protected] [email protected] [email protected] Lawrence Livermore National Lawrence Livermore National Lawrence Livermore National Laboratory Laboratory Laboratory Livermore, USA Livermore, USA Livermore, USA Keita Iwabuchi Kai Wu Dong Li [email protected] [email protected] [email protected] Lawrence Livermore National University of California, Merced University of California, Merced Laboratory Merced, USA Merced, USA Livermore, USA Roger Pearce Maya Gokhale [email protected] [email protected] Lawrence Livermore National Lawrence Livermore National Laboratory Laboratory Livermore, USA Livermore, USA ABSTRACT KEYWORDS Leadership supercomputers feature a diversity of storage, memory mapping, memmap, page fault, user-space paging, from node-local persistent memory and NVMe SSDs to userfaultfd, page management network-interconnected ash memory and HDD. Memory Reference Format: mapping les on dierent tiers of storage provides a uniform Ivy B. Peng, Marty McFadden, Eric Green, Keita Iwabuchi, Kai Wu, interface in applications. However, system-wide services like Dong Li, Roger Pearce, and Maya Gokhale. 2019. UMap : Enabling mmap are optimized for generality and lack exibility for Application-driven Optimizations for Page Management. enabling application-specic optimizations. In this work, we present UMap to enable user-space page management that 1 INTRODUCTION can be easily adapted to access patterns in applications and Recently, leadership supercomputers provide enormous stor- storage characteristics. UMap uses the userfaultfd mecha- age resources to cope with expanding data sets in applica- nism to handle page faults in multi-threaded applications tions. The storage resources come in a hybrid format for eciently. By providing a data object abstraction layer, UMap balanced cost and performance [9, 11, 13]. Fast and small is extensible to support various backing stores. The design of storage, which is implemented using advanced technologies UMap supports dynamic load balancing and I/O decoupling like persistent memory and NVMe SSDs, often co-locate with for scalable performance. UMap also uses application hints computing units inside compute node. Storage with massive to improve the selection of caching, prefetching, and eviction capacity, on the other hand, uses cost-eective technologies policies. We evaluate UMap in ve benchmarks and real ap- like HDD and is interconnected to compute nodes through plications on two systems. Our results show that leveraging the network. In between, burst buers use fast memory tech- application knowledge for page management could substan- nologies and are accessible through the network. Memory tially improve performance. On average, UMap achieved 1.25 mapping provides a uniform interface to access les on dif- to 2.5 times improvement using the adapted congurations ferent types of storage as if to dynamically allocated memory. compared to the system service. For instance, out-of-core data analytic workloads often need to process large datasets that exceed the memory capacity of a compute node [17]. Using memory mapping to access MCHPC’19, Denver, USA these datasets shift the burden of paging, prefetching, and 2019. ACM ISBN 978-x-xxxx-xxxx-x/YY/MM. caching data between storage and memory to the operating systems. MCHPC’19, Denver, USA Peng et al. Currently, operating systems provide the mmap system SSD, applications achieved 1.25 to 2.5 times improvement call to map les or devices into memory. This system service compared to the standard system service. On the Intel testbed performs well in loading dynamic libraries and could also with network-interconnected HDD, UMap brings the per- support out-of-core execution. However, as a system-level formance of the asteroid detection application close to that service, it has to be tuned for performance reliability and uses local SSD for 500 GB data sets. In summary, our main consistency over a broad range of workloads. Therefore, it contributions are as follows: may reduce opportunities in optimizing performance based We propose an open-source library1, called UMap that • on application characteristics. Moreover, backing stores on leverages lightweight userfaultfd mechanism to enable dierent storage exhibit distinctive performance character- application-driven page management. istics. Consequently, congurations tuned for one type of We describe the design of UMap for achieving scalable • storage will need to be adjusted when mapping on another performance in multi-threaded data-intensive applica- type of storage. In this work, we provide UMap to enable tions. application-specic optimizations for page management in We demonstrate ve use cases of UMap and show that • memory mapping various backing stores. UMap is highly enabling congurable page size is essential for perfor- congurable to adapt user-space paging to suit application mance tuning in data-intensive applications. needs. It facilitates application control on caching, prefetch- UMap improves the performance of tested applications • ing, and eviction policies with minimal porting eorts from by 1.25 to 2.5 times compared to the standard mmap the programmer. As a user-level solution, UMap connes system service. changes within an application without impacting other ap- plications sharing the platform, which is unachievable in 2 BACKGROUND AND MOTIVATION system-level approaches. In this section, we introduce memory mapping, prospective We prioritize four design choices for UMap based on sur- benets from user-space page management, and the enabling veying realistic use cases. First, we choose to implement mechanism userfaultfd. UMap as a user-level library so that it can maintain compati- bility with the fast-moving Linux kernel without the need 2.1 Memory Mapping to track and modify for frequent kernel updates. Also, we Memory mapping links images and les in persistent storage employ the recent userfaultfd [7] mechanism, other than to the virtual address space of a process. The operating sys- the signal handling + callback function approach to reduce tem employs demand paging to bring only accessed virtual overhead and performance variance in multi-threaded appli- pages into physical memory because virtual memory can be cations. Third, we target an adaptive solution that sustains much larger than physical memory. An access to memory- performance even at high concurrency for data-intensive mapped regions triggers a page fault if no page table entry applications, which often employ a large number of threads (PTE) is present for the accessed page. When such a page for hiding data access latency. Our design pays particular fault is raised, the operating system resolves it by copying in consideration on load imbalance among service threads to the physical data page from storage to the in-memory page improve the utilization of shared resources even when data cache. accesses to pages are skewed. UMap dynamically balances Common strategies for optimizing memory mapping in workloads among all service threads to eliminate bottleneck the operating systems include page cache, read-ahead, and on serving hot pages. Finally, for exible and portable tun- madvise hints. The page cache is used to keep frequently used ing on dierent computing systems, UMap provides both pages in memory while less important pages may need to API and environmental controls to enable congurable page be evicted from memory to make room for newly requested sizes, eviction strategy, application-specic prefetching, and pages. Least Recently Used (LRU) policy is commonly used detailed diagnosis information to the programmer. for selecting pages to be evicted. The operating system may We evaluate the eectiveness of UMap in ve use cases, proactively ush dirty pages, i.e., modied pages in the page including two data-intensive benchmarks, i.e., a synthetic cache, into storage when the ratio of dirty page exceeds a sort benchmark and a breadth-rst search (BFS) kernel, and threshold value [19]. Read-ahead preloads pages into physi- three real applications, i.e., Lrzip [8], N-Store database [2], cal memory to avoid the overhead associated with page fault and an asteroid detection application that processes massive handling, TLB misses and user-to-kernel mode transition. data sets from telescopes. We conduct out-of-core experi- Finally, the madvise interface takes hints to allow the op- ments on two systems with node-local SSD and network- erating system to make informed decisions for managing interconnected HDD storage. Our results show that UMap pages. can enable exible user-space page management in data- intensive applications. On the AMD testbed with local NVMe 1UMAP v2.0.0 https://github.com/LLNL/umap. UMap : Enabling Application-driven Optimizations for Page Management MCHPC’19, Denver, USA 2.2 User-space Page Management Application User-space page management uses application threads to resolve page faults and manage virtual memory in the back- ground as dened by the application. The userfaultfd is a Virtual Address Space lightweight mechanism to enable user-space paging com- UMap Physical pages pared to the traditional SIGSEGV signal and callback func- Page faults ... tion [7]. Applications register address ranges to be manged in user-space, and specify the type of events, e.g., page faults Filler 0 Filler 1 Filler 2 Filler 3 and events in un-cooperative mode, to be tracked. Page faults Store Store in the address ranges are delivered
Load more

Recommended publications
  • Proceedings 2005
    LAC2005 Proceedings 3rd International Linux Audio Conference April 21 – 24, 2005 ZKM | Zentrum fur¨ Kunst und Medientechnologie Karlsruhe, Germany Published by ZKM | Zentrum fur¨ Kunst und Medientechnologie Karlsruhe, Germany April, 2005 All copyright remains with the authors www.zkm.de/lac/2005 Content Preface ............................................ ............................5 Staff ............................................... ............................6 Thursday, April 21, 2005 – Lecture Hall 11:45 AM Peter Brinkmann MidiKinesis – MIDI controllers for (almost) any purpose . ....................9 01:30 PM Victor Lazzarini Extensions to the Csound Language: from User-Defined to Plugin Opcodes and Beyond ............................. .....................13 02:15 PM Albert Gr¨af Q: A Functional Programming Language for Multimedia Applications .........21 03:00 PM St´ephane Letz, Dominique Fober and Yann Orlarey jackdmp: Jack server for multi-processor machines . ......................29 03:45 PM John ffitch On The Design of Csound5 ............................... .....................37 04:30 PM Pau Arum´ıand Xavier Amatriain CLAM, an Object Oriented Framework for Audio and Music . .............43 Friday, April 22, 2005 – Lecture Hall 11:00 AM Ivica Ico Bukvic “Made in Linux” – The Next Step .......................... ..................51 11:45 AM Christoph Eckert Linux Audio Usability Issues .......................... ........................57 01:30 PM Marije Baalman Updates of the WONDER software interface for using Wave Field Synthesis . 69 02:15 PM Georg B¨onn Development of a Composer’s Sketchbook ................. ....................73 Saturday, April 23, 2005 – Lecture Hall 11:00 AM J¨urgen Reuter SoundPaint – Painting Music ........................... ......................79 11:45 AM Michael Sch¨uepp, Rene Widtmann, Rolf “Day” Koch and Klaus Buchheim System design for audio record and playback with a computer using FireWire . 87 01:30 PM John ffitch and Tom Natt Recording all Output from a Student Radio Station .
    [Show full text]
  • System Calls System Calls
    System calls We will investigate several issues related to system calls. Read chapter 12 of the book Linux system call categories file management process management error handling note that these categories are loosely defined and much is behind included, e.g. communication. Why? 1 System calls File management system call hierarchy you may not see some topics as part of “file management”, e.g., sockets 2 System calls Process management system call hierarchy 3 System calls Error handling hierarchy 4 Error Handling Anything can fail! System calls are no exception Try to read a file that does not exist! Error number: errno every process contains a global variable errno errno is set to 0 when process is created when error occurs errno is set to a specific code associated with the error cause trying to open file that does not exist sets errno to 2 5 Error Handling error constants are defined in errno.h here are the first few of errno.h on OS X 10.6.4 #define EPERM 1 /* Operation not permitted */ #define ENOENT 2 /* No such file or directory */ #define ESRCH 3 /* No such process */ #define EINTR 4 /* Interrupted system call */ #define EIO 5 /* Input/output error */ #define ENXIO 6 /* Device not configured */ #define E2BIG 7 /* Argument list too long */ #define ENOEXEC 8 /* Exec format error */ #define EBADF 9 /* Bad file descriptor */ #define ECHILD 10 /* No child processes */ #define EDEADLK 11 /* Resource deadlock avoided */ 6 Error Handling common mistake for displaying errno from Linux errno man page: 7 Error Handling Description of the perror () system call.
    [Show full text]
  • Dell Update Packages for Linux Operating Systems User's Guide
    Dell™ Update Packages for Linux Operating Systems User’s Guide Notes and Cautions NOTE: A NOTE indicates important information that helps you make better use of your computer. CAUTION: A CAUTION indicates potential damage to hardware or loss of data if instructions are not followed. ____________________ Information in this document is subject to change without notice. © 2009 Dell Inc. All rights reserved. Reproduction of these materials in any manner whatsoever without the written permission of Dell Inc. is strictly forbidden. Trademarks used in this text: Dell, the DELL logo, and OpenManage are trademarks of Dell Inc.; Microsoft and Windows are either trademarks or registered trademarks of Microsoft Corporation in the United States and/or other countries; Intel is a registered trademark of Intel Corporation in the United States and other countries; Red Hat and Red Hat Enterprise Linux are registered trademarks of Red Hat, Inc. in the United States and other countries; SUSE is a registered trademark of Novell, Inc. in the United States and other countries; VMware and ESX Server are registered trademarks or trademarks of VMware, Inc. in the United States and/or other jurisdictions; Citrix and XenServer are either trademarks or registered trademarks of Citrix Systems, Inc. in the United States and/or other countries. Other trademarks and trade names may be used in this document to refer to either the entities claiming the marks and names or their products. Dell Inc. disclaims any proprietary interest in trademarks and trade names other than its own. April 2009 Contents 1 Getting Started With Dell Update Packages . 7 Overview .
    [Show full text]
  • Inside the Linux 2.6 Completely Fair Scheduler
    Inside the Linux 2.6 Completely Fair Scheduler http://www.ibm.com/developerworks/library/l-completely-fair-scheduler/ developerWorks Technical topics Linux Technical library Inside the Linux 2.6 Completely Fair Scheduler Providing fair access to CPUs since 2.6.23 The task scheduler is a key part of any operating system, and Linux® continues to evolve and innovate in this area. In kernel 2.6.23, the Completely Fair Scheduler (CFS) was introduced. This scheduler, instead of relying on run queues, uses a red-black tree implementation for task management. Explore the ideas behind CFS, its implementation, and advantages over the prior O(1) scheduler. M. Tim Jones is an embedded firmware architect and the author of Artificial Intelligence: A Systems Approach, GNU/Linux Application Programming (now in its second edition), AI Application Programming (in its second edition), and BSD Sockets Programming from a Multilanguage Perspective. His engineering background ranges from the development of kernels for geosynchronous spacecraft to embedded systems architecture and networking protocols development. Tim is a Consultant Engineer for Emulex Corp. in Longmont, Colorado. 15 December 2009 Also available in Japanese The Linux scheduler is an interesting study in competing pressures. On one side are the use models in which Linux is applied. Although Linux was originally developed as a desktop operating system experiment, you'll now find it on servers, tiny embedded devices, mainframes, and supercomputers. Not surprisingly, the scheduling loads for these domains differ. On the other side are the technological advances made in the platform, including architectures (multiprocessing, symmetric multithreading, non-uniform memory access [NUMA]) and virtualization.
    [Show full text]
  • Thread Scheduling in Multi-Core Operating Systems Redha Gouicem
    Thread Scheduling in Multi-core Operating Systems Redha Gouicem To cite this version: Redha Gouicem. Thread Scheduling in Multi-core Operating Systems. Computer Science [cs]. Sor- bonne Université, 2020. English. tel-02977242 HAL Id: tel-02977242 https://hal.archives-ouvertes.fr/tel-02977242 Submitted on 24 Oct 2020 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. Ph.D thesis in Computer Science Thread Scheduling in Multi-core Operating Systems How to Understand, Improve and Fix your Scheduler Redha GOUICEM Sorbonne Université Laboratoire d’Informatique de Paris 6 Inria Whisper Team PH.D.DEFENSE: 23 October 2020, Paris, France JURYMEMBERS: Mr. Pascal Felber, Full Professor, Université de Neuchâtel Reviewer Mr. Vivien Quéma, Full Professor, Grenoble INP (ENSIMAG) Reviewer Mr. Rachid Guerraoui, Full Professor, École Polytechnique Fédérale de Lausanne Examiner Ms. Karine Heydemann, Associate Professor, Sorbonne Université Examiner Mr. Etienne Rivière, Full Professor, University of Louvain Examiner Mr. Gilles Muller, Senior Research Scientist, Inria Advisor Mr. Julien Sopena, Associate Professor, Sorbonne Université Advisor ABSTRACT In this thesis, we address the problem of schedulers for multi-core architectures from several perspectives: design (simplicity and correct- ness), performance improvement and the development of application- specific schedulers.
    [Show full text]
  • Package Manager: the Core of a GNU/Linux Distribution
    Package Manager: The Core of a GNU/Linux Distribution by Andrey Falko A Thesis submitted to the Faculty in partial fulfillment of the requirements for the BACHELOR OF ARTS Accepted Paul Shields, Thesis Adviser Allen Altman, Second Reader Christopher K. Callanan, Third Reader Mary B. Marcy, Provost and Vice President Simon’s Rock College Great Barrington, Massachusetts 2007 Abstract Package Manager: The Core of a GNU/Linux Distribution by Andrey Falko As GNU/Linux operating systems have been growing in popularity, their size has also been growing. To deal with this growth people created special programs to organize the software available for GNU/Linux users. These programs are called package managers. There exists a very wide variety of package managers, all offering their own benefits and deficiencies. This thesis explores all of the major aspects of package management in the GNU/Linux world. It covers what it is like to work without package managers, primitive package man- agement techniques, and modern package management schemes. The thesis presents the creation of a package manager called Vestigium. The creation of Vestigium is an attempt to automate the handling of file collisions between packages, provide a seamless interface for installing both binary packages and packages built from source, and to allow per package optimization capabilities. Some of the features Vestigium is built to have are lacking in current package managers. No current package manager contains all the features which Vestigium is built to have. Additionally, the thesis explains the problems that developers face in maintaining their respective package managers. i Acknowledgments I thank my thesis committee members, Paul Shields, Chris Callanan, and Allen Altman for being patient with my error-ridden drafts.
    [Show full text]
  • Machine Learning for Load Balancing in the Linux Kernel Jingde Chen Subho S
    Machine Learning for Load Balancing in the Linux Kernel Jingde Chen Subho S. Banerjee [email protected] [email protected] University of Illinois at Urbana-Champaign University of Illinois at Urbana-Champaign Zbigniew T. Kalbarczyk Ravishankar K. Iyer [email protected] [email protected] University of Illinois at Urbana-Champaign University of Illinois at Urbana-Champaign Abstract ACM Reference Format: The OS load balancing algorithm governs the performance Jingde Chen, Subho S. Banerjee, Zbigniew T. Kalbarczyk, and Rav- gains provided by a multiprocessor computer system. The ishankar K. Iyer. 2020. Machine Learning for Load Balancing in Linux’s Completely Fair Scheduler (CFS) scheduler tracks the Linux Kernel. In 11th ACM SIGOPS Asia-Pacific Workshop on Systems (APSys ’20), August 24–25, 2020, Tsukuba, Japan. ACM, New process loads by average CPU utilization to balance work- York, NY, USA, 8 pages. https://doi.org/10.1145/3409963.3410492 load between processor cores. That approach maximizes the utilization of processing time but overlooks the con- tention for lower-level hardware resources. In servers run- 1 Introduction ning compute-intensive workloads, an imbalanced need for The Linux scheduler, called the Completely Fair Scheduler limited computing resources hinders execution performance. (CFS) algorithm, enforces a fair scheduling policy that aims to This paper solves the above problem using a machine learn- allow all tasks to be executed with a fair quota of processing ing (ML)-based resource-aware load balancer. We describe time. CFS allocates CPU time appropriately among the avail- (1) low-overhead methods for collecting training data; (2) an able hardware threads to allow processes to make forward ML model based on a multi-layer perceptron model that imi- progress.
    [Show full text]
  • Linux CPU Schedulers: CFS and Muqss Comparison
    Umeå University 2021-06-14 Bachelor’s Thesis In Computing Science Linux CPU Schedulers: CFS and MuQSS Comparison Name David Shakoori Gustafsson Supervisor Anna Jonsson Examinator Ola Ringdahl CFS and MuQSS Comparison Abstract The goal of this thesis is to compare two process schedulers for the Linux operating system. In order to provide a responsive and interactive user ex- perience, an efficient process scheduling algorithm is important. This thesis seeks to explain the potential performance differences by analysing the sched- ulers’ respective designs. The two schedulers that are tested and compared are Con Kolivas’s MuQSS and Linux’s default scheduler, CFS. They are tested with respect to three main aspects: latency, turn-around time and interactivity. Latency is tested by using benchmarking software, the turn- around time by timing software compilation, and interactivity by measuring video frame drop percentages under various background loads. These tests are performed on a desktop PC running Linux OpenSUSE Leap 15.2, using kernel version 5.11.18. The test results show that CFS manages to keep a generally lower latency, while turn-around times differs little between the two. Running the turn-around time test’s compilation using a single process gives MuQSS a small advantage, while dividing the compilation evenly among the available logical cores yields little difference. However, CFS clearly outper- forms MuQSS in the interactivity test, where it manages to keep frame drop percentages considerably lower under each tested background load. As is apparent by the results, Linux’s current default scheduler provides a more responsive and interactive experience within the testing conditions, than the alternative MuQSS.
    [Show full text]
  • Linux-Cookbook.Pdf
    LINUX COOKBOOK ™ Other Linux resources from O’Reilly Related titles Linux Device Drivers Exploring the JDS Linux Linux in a Nutshell Desktop Running Linux Learning Red Hat Enterprise Building Embedded Linux Linux and Fedora Systems Linux Pocket Guide Linux Security Cookbook Understanding the Linux Kernel Linux Books linux.oreilly.com is a complete catalog of O’Reilly’s books on Resource Center Linux and Unix and related technologies, including sample chapters and code examples. ONLamp.com is the premier site for the open source web plat- form: Linux, Apache, MySQL, and either Perl, Python, or PHP. Conferences O’Reilly brings diverse innovators together to nurture the ideas that spark revolutionary industries. We specialize in document- ing the latest tools and systems, translating the innovator’s knowledge into useful skills for those in the trenches. Visit conferences.oreilly.com for our upcoming events. Safari Bookshelf (safari.oreilly.com) is the premier online refer- ence library for programmers and IT professionals. Conduct searches across more than 1,000 books. Subscribers can zero in on answers to time-critical questions in a matter of seconds. Read the books on your Bookshelf from cover to cover or sim- ply flip to the page you need. Try it today with a free trial. LINUX COOKBOOK ™ Carla Schroder Beijing • Cambridge • Farnham • Köln • Paris • Sebastopol • Taipei • Tokyo Linux Cookbook™ by Carla Schroder Copyright © 2005 O’Reilly Media, Inc. All rights reserved. Printed in the United States of America. Published by O’Reilly Media, Inc., 1005 Gravenstein Highway North, Sebastopol, CA 95472. O’Reilly books may be purchased for educational, business, or sales promotional use.
    [Show full text]
  • Evolutions in Recent Hardware and Applications to Virtualization Clementine Maurice
    Information leakage on shared hardware : evolutions in recent hardware and applications to virtualization Clementine Maurice To cite this version: Clementine Maurice. Information leakage on shared hardware : evolutions in recent hardware and applications to virtualization. Cryptography and Security [cs.CR]. Télécom ParisTech, 2015. English. NNT : 2015ENST0070. tel-01439872 HAL Id: tel-01439872 https://pastel.archives-ouvertes.fr/tel-01439872 Submitted on 18 Jan 2017 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. 2015-ENST-0070 EDITE - ED 130 Doctorat ParisTech THÈSE pour obtenir le grade de docteur délivré par TELECOM ParisTech Spécialité « Informatique » présentée et soutenue publiquement par Clémentine MAURICE le 28 octobre 2015 Fuites d’information dans les processeurs récents et applications à la virtualisation Directeurs de thèse : Aurélien FRANCILLON Christoph NEUMANN Co-encadrement de la thèse : Olivier HEEN Jury M. Thomas JENSEN, Directeur de recherche, Inria Rennes, France Président de jury M. Gildas AVOINE, Professeur, IRISA et INSA Rennes,
    [Show full text]
  • Cross-Layer Resource Control and Scheduling for Improving Interactivity in Android
    SOFTWARE—PRACTICE AND EXPERIENCE Softw. Pract. Exper. 2014; 00:1–22 Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/spe Cross-Layer Resource Control and Scheduling for Improving Interactivity in Android ∗ † Sungju Huh1, Jonghun Yoo2 and Seongsoo Hong1,2,3, , 1Department of Transdisciplinary Studies, Graduate School of Convergence Science and Technology, Seoul National University, Suwon-si, Gyeonggi-do, Republic of Korea 2Department of Electrical and Computer Engineering, Seoul National University, Seoul, Republic of Korea 3Advanced Institutes of Convergence Technology, Suwon-si, Gyeonggi-do, Republic of Korea SUMMARY Android smartphones are often reported to suffer from sluggish user interactions due to poor interactivity. This is partly because Android and its task scheduler, the completely fair scheduler (CFS), may incur perceptibly long response time to user-interactive tasks. Particularly, the Android framework cannot systemically favor user-interactive tasks over other background tasks since it does not distinguish between them. Furthermore, user-interactive tasks can suffer from high dispatch latency due to the non-preemptive nature of CFS. To address these problems, this paper presents framework-assisted task characterization and virtual time-based CFS. The former is a cross-layer resource control mechanism between the Android framework and the underlying Linux kernel. It identifies user-interactive tasks at the framework-level, by using the notion of a user-interactive task chain. It then enables the kernel scheduler to selectively promote the priorities of worker tasks appearing in the task chain to reduce the preemption latency. The latter is a cross-layer refinement of CFS in terms of interactivity. It allows a task to be preempted at every predefined period.
    [Show full text]
  • Linux Internals
    LINUX INTERNALS Peter Chubb and Etienne Le Sueur [email protected] A LITTLE BIT OF HISTORY • Ken Thompson and Dennis Ritchie in 1967–70 • USG and BSD • John Lions 1976–95 • Andrew Tanenbaum 1987 • Linux Torvalds 1991 NICTA Copyright c 2011 From Imagination to Impact 2 The history of UNIX-like operating systems is a history of people being dissatisfied with what they have and wanting to do some- thing better. It started when Ken Thompson got bored with MUL- TICS and wanted to write a computer game (Space Travel). He found a disused PDP-7, and wrote an interactive operating sys- tem to run his game. The main contribution at this point was the simple file-system abstraction. (Ritchie 1984) Other people found it interesting enough to want to port it to other systems, which led to the first major rewrite — from assembly to C. In some ways UNIX was the first successfully portable OS. After Ritchie & Thompson (1974) was published, AT&T became aware of a growing market for UNIX. They wanted to discourage it: it was common for AT&T salesmen to say, ‘Here’s what you get: A whole lot of tapes, and an invoice for $10 000’. Fortunately educational licences were (almost) free, and universities around the world took up UNIX as the basis for teaching and research. The University of California at Berkeley was one of those univer- NICTA Copyright c 2011 From Imagination to Impact 2-1 sities. In 1977, Bill Joy (a postgrad) put together and released the first Berkeley Software Distribution — in this instance, the main additions were a pascal compiler and Bill Joy’s ex editor.
    [Show full text]