DOCSLIB.ORG

X-Containers: Breaking Down Barriers to Improve

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
X-Containers: Breaking Down Barriers to Improve Session: Cloud II ASPLOS’19, April 13–17, 2019, Providence, RI, USA X-Containers: Breaking Down Barriers to Improve Performance and Isolation of Cloud-Native Containers Zhiming Shen Zhen Sun Gur-Eyal Sela∗ Cornell University Cornell University University of California, Berkeley Eugene Bagdasaryan Christina Delimitrou Robbert Van Renesse Cornell University Cornell University Cornell University Hakim Weatherspoon Cornell University Abstract CCS Concepts • Security and privacy → Virtualiza- “Cloud-native” container platforms, such as Kubernetes, have tion and security; • Software and its engineering → become an integral part of production cloud environments. Operating systems. One of the principles in designing cloud-native applica- Keywords Containers; X-Containers; Cloud-Native; Li- tions is called Single Concern Principle, which suggests that brary OS; exokernel each container should handle a single responsibility well. In this paper, we propose X-Containers as a new security ACM Reference Format: paradigm for isolating single-concerned cloud-native con- Zhiming Shen, Zhen Sun, Gur-Eyal Sela, Eugene Bagdasaryan, tainers. Each container is run with a Library OS (LibOS) Christina Delimitrou, Robbert Van Renesse, and Hakim Weath- that supports multi-processing for concurrency and compat- erspoon. 2019. X-Containers: Breaking Down Barriers to Improve Performance and Isolation of Cloud-Native Containers. In 2019 Ar- ibility. A minimal exokernel ensures strong isolation with chitectural Support for Programming Languages and Operating Sys- small kernel attack surface. We show an implementation tems (ASPLOS ’19), April 13–17, 2019, Providence, RI, USA. ACM, New of the X-Containers architecture that leverages Xen para- York, NY, USA, 15 pages. https://doi.org/10.1145/3297858.3304016 virtualization (PV) to turn Linux kernel into a LibOS. Do- ing so results in a highly efficient LibOS platform that does 1 Introduction not require hardware-assisted virtualization, improves inter- container isolation, and supports binary compatibility and An important recent trend in cloud computing is the rise of multi-processing. By eliminating some security barriers such “cloud-native” container platforms, such as Kubernetes [38], as seccomp and Meltdown patch, X-Containers have up to which have become an integral part of production envi- 27× higher raw system call throughput compared to Docker ronments. Such platforms support applications designed containers, while also achieving competitive or superior per- specifically for cloud infrastructures that consist of loosely- formance on various benchmarks compared to recent con- coupled microservices [62] running in containers, enabling tainer platforms such as Google’s gVisor and Intel’s Clear automatic orchestration and agile DevOps practices [33]. In Containers. cloud-native platforms, container design is similar to object design in object-oriented (OO) software systems: each con- tainer should have a single responsibility and handle that ∗Work conducted at Cornell University. responsibility well [39]. By focusing on a single concern, cloud-native containers are easier to scale horizontally, and Permission to make digital or hard copies of all or part of this work for replace, reuse, and upgrade transparently. Similar to the Sin- personal or classroom use is granted without fee provided that copies gle Responsibility Principle in OO-languages, this has been are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights termed the “Single Concern Principle” [51], and is recom- for components of this work owned by others than the author(s) must mended by Docker [8]. be honored. Abstracting with credit is permitted. To copy otherwise, or Running multiple containers on the same host does not republish, to post on servers or to redistribute to lists, requires prior specific come without problems. From a security perspective, if one permission and/or a fee. Request permissions from [email protected]. container is compromised, all containers on the same Op- ASPLOS ’19, April 13–17, 2019, Providence, RI, USA erating System (OS) kernel are put under risk. Due to the © 2019 Copyright held by the owner/author(s). Publication rights licensed to ACM. concern of application isolation, containers are generally ACM ISBN 978-1-4503-6240-5/19/04...$15.00 not allowed to install their own kernel modules, a limitation https://doi.org/10.1145/3297858.3304016 for applications that require kernel customization. Nor can 121 Session: Cloud II ASPLOS’19, April 13–17, 2019, Providence, RI, USA the OS kernel be easily tuned and optimized for a particular seccomp filters and Meltdown [57] patch, X-Containers have container since it is shared by other containers. up to 27× higher raw system call throughput compared to There have been several proposals to address the is- native Docker containers running in the cloud. X-Containers sue of container isolation. Hypervisor-based container run- also achieve competitive or superior performance compared times [17], such as Clear Containers [15], Kata Contain- to recent container platforms such as gVisor and Clear Con- ers [16], and Hyper Containers [13], wrap containers with tainers, as well as other LibOSes like Unikernel and Graphene a dedicated OS kernel running in a virtual machine (VM). on various benchmarks. These platforms require hardware-assisted virtualization sup- The X-Container architecture, however, also imposes sev- port to reduce the overhead of adding another layer of indi- eral limitations. For example, the change in the threat model rection. However, many public and private clouds, including makes it unsuitable for running some containers that still Amazon EC2, do not support nested hardware virtualization. require process and kernel isolation. Due to the require- Even in clouds like Google Compute Engine where nested ment of running a LibOS with each container, X-Containers hardware virtualization is enabled, its performance overhead take longer time to boot and have bigger memory footprint. is high (Section 5 and [5]). LightVM [60] wraps a container X-Containers also face challenges of page table operation in a paravirtualized Xen instance without hardware virtu- efficiency and dynamic memory management. We discuss alization support. Unfortunately, it introduces a significant these limitations in the paper. performance penalty in x86-64 platforms (Section 4.1 and This paper makes the following contributions: 5). Finally, Google gVisor [12] is a user-space kernel written • We present X-Containers, a new exokernel-inspired in Go that supports container runtime sandboxing, but it container architecture that is designed specifically for only offers limited system call compatibility55 [ ] and incurs single-concerned cloud-native applications. We dis- significant performance overheads (Section 5). cuss the new threat model and the trade-offs it intro- The trend of running a single application in its own VM duces, the advantages and limitations of the proposed for enhanced security has led to a renewed interest in Lib- design, including those related to running unmodified OSes, as suggested by the Unikernel [58] model. LibOSes applications in X-Containers. avoid the overhead of security isolation between the appli- • We demonstrate how the Xen paravirtualization ar- cation and the OS, and allow each LibOS to be carefully chitecture and the Linux kernel can be turned into a optimized for the application at hand. Designing a container secure and efficient LibOS platform that supports both architecture inspired by the exokernel+LibOS [43] model can binary compatibility and multi-processing. improve both container isolation and performance. However, • We present a technology for automatically changing existing LibOSes, such as MirageOS [58], Graphene [69], system calls into function calls to optimize applications and OSv [53], lack features like full binary compatibility or running on a LibOS. multi-processing support. This makes porting containerized • We evaluate the efficacy of X-Containers against applications very challenging. Docker, gVisor, Clear Container, and other LibOSes In this paper, we propose a new LibOS platform called (Unikernel and Graphene), and demonstrate competi- X-Containers that improves container isolation without re- tive or superior performance. quiring hardware virtualization support. An X-Container can support one or more user processes that all run at the same privilege level as the LibOS. Different processes inside an 2 X-Containers as a New Security X-Container still have their own address spaces for resource Paradigm management and compatibility, but they no longer provide secure isolation from one another; in this new security para- 2.1 Single-Concerned Containers digm processes are used for concurrency, while X-Containers Cloud-native applications are designed to fully exploit the provide isolation between containers. We show an implemen- potential of cloud infrastructures. Although legacy applica- tation of the X-Containers architecture that leverages Xen’s tions can be packaged in containers and run in a cloud, these paravirtualization (PV) architecture [32] and turns the Linux applications cannot take full advantage of the automated kernel into a LibOS that supports both binary compatibility deployment, scaling, and orchestration offered by systems and multi-processing. like Kubernetes, which are designed for single-concerned Without hardware virtualization
Load more

Recommended publications
  • Benchmarking, Analysis, and Optimization of Serverless Function Snapshots
    Benchmarking, Analysis, and Optimization of Serverless Function Snapshots Dmitrii Ustiugov∗ Plamen Petrov Marios Kogias† University of Edinburgh University of Edinburgh Microsoft Research United Kingdom United Kingdom United Kingdom Edouard Bugnion Boris Grot EPFL University of Edinburgh Switzerland United Kingdom ABSTRACT CCS CONCEPTS Serverless computing has seen rapid adoption due to its high scala- • Computer systems organization ! Cloud computing; • In- bility and flexible, pay-as-you-go billing model. In serverless, de- formation systems ! Computing platforms; Data centers; • velopers structure their services as a collection of functions, spo- Software and its engineering ! n-tier architectures. radically invoked by various events like clicks. High inter-arrival time variability of function invocations motivates the providers KEYWORDS to start new function instances upon each invocation, leading to cloud computing, datacenters, serverless, virtualization, snapshots significant cold-start delays that degrade user experience. To reduce ACM Reference Format: cold-start latency, the industry has turned to snapshotting, whereby Dmitrii Ustiugov, Plamen Petrov, Marios Kogias, Edouard Bugnion, and Boris an image of a fully-booted function is stored on disk, enabling a Grot. 2021. Benchmarking, Analysis, and Optimization of Serverless Func- faster invocation compared to booting a function from scratch. tion Snapshots . In Proceedings of the 26th ACM International Conference on This work introduces vHive, an open-source framework for Architectural Support for Programming Languages and Operating Systems serverless experimentation with the goal of enabling researchers (ASPLOS ’21), April 19–23, 2021, Virtual, USA. ACM, New York, NY, USA, to study and innovate across the entire serverless stack. Using 14 pages. https://doi.org/10.1145/3445814.3446714 vHive, we characterize a state-of-the-art snapshot-based serverless infrastructure, based on industry-leading Containerd orchestra- 1 INTRODUCTION tion framework and Firecracker hypervisor technologies.
    [Show full text]
  • The Next Generation Cloud: the Rise of the Unikernel
    The Next Generation Cloud: The Rise of the Unikernel A Xen Project publication April 2015 xenproject.org Docker and Linux container technologies dominate headlines today as a powerful, easy way to package applications, especially as cloud computing becomes more mainstream. While still a work-in-progress, they offer a simple, clean and lean way to distribute application workloads. With enthusiasm continuing to grow for container innovations, a related technology called unikernels is also beginning to attract attention. Known also for their ability to cleanly separate functionality at the component level, unikernels are developing a variety of new approaches to deploy cloud services. Traditional operating systems run multiple applications on a single machine, managing resources and isolating applications from one another. A unikernel runs a single application on a single virtual machine, relying instead on the hypervisor to isolate those virtual machines. Unikernels are constructed by using “library operating systems,” from which the developer selects only the minimal set of services required for an application to run. These sealed, fixed-purpose images run directly on a hypervisor without an intervening guest OS such as Linux. As well as improving upon container technologies, unikernels are also able to deliver impressive flexibility, speed and versatility for cross-platform environments, big data analytics and scale-out cloud computing. Like container-based solutions, this technology fulfills the promise of easy deployment, but unikernels also offer an extremely tiny, specialized runtime footprint that is much less vulnerable to attack. There are several up-and-coming open source projects to watch this year, including ClickOS, Clive, HaLVM, LING, MirageOS, Rump Kernels and OSv among others, with each of them placing emphasis on a different aspect of the unikernel approach.
    [Show full text]
  • Qualifikationsprofil #10309
    QUALIFIKATIONSPROFIL #10309 ALLGEMEINE DATEN Geburtsjahr: 1972 Ausbildung: Abitur Diplom, Informatik, (TU, Kaiserslautern) Fremdsprachen: Englisch, Französisch Spezialgebiete: Kubernetes KENNTNISSE Tools Active Directory Apache Application Case CATIA CVS Eclipse Exchange Framework GUI Innovator ITIL J2EE JMS LDAP Lotus Notes make MS Exchange MS Outlook MS-Exchange MS-Office MS-Visual Studio NetBeans OSGI RACF SAS sendmail Together Turbine UML VMWare .NET ADS ANT ASP ASP.NET Flash GEnie IDES Image Intellij IDEA IPC Jackson JBOSS Lex MS-Visio ODBC Oracle Application Server OWL PGP SPSS SQS TesserAct Tivoli Toolbook Total Transform Visio Weblogic WebSphere YACC Tätigkeiten Administration Analyse Beratung Design Dokumentation KI Konzeption Optimierung Support Vertrieb Sprachen Ajax Basic C C# C++ Cobol Delphi ETL Fortran Java JavaScript Natural Perl PHP PL/I PL-SQL Python SAL Smalltalk SQL ABAP Atlas Clips Delta FOCUS HTML Nomad Pascal SPL Spring TAL XML Detaillierte Komponenten AM BI FS-BA MDM PDM PM BW CO FI LO PP Datenbanken Approach IBM Microsoft Object Store Oracle Progress Sybase DMS ISAM JDBC mySQL DC/Netzwerke ATM DDS Gateway HBCI Hub Internet Intranet OpenSSL SSL VPN Asynchronous CISCO Router DNS DSL Firewall Gateways HTTP RFC Router Samba Sockets Switches Finance Business Intelligence Excel Konsolidierung Management Projektleiter Reporting Testing Wertpapiere Einkauf CAD Systeme CATIA V5 sonstige Hardware Digital HP PC Scanner Siemens Spark Teradata Bus FileNet NeXT SUN Switching Tools, Methoden Docker Go Kubernetes Rational RUP
    [Show full text]
  • A Linux in Unikernel Clothing Lupine
    A Linux in Unikernel Clothing Hsuan-Chi Kuo+, Dan Williams*, Ricardo Koller* and Sibin Mohan+ +University of Illinois at Urbana-Champaign *IBM Research Lupine Unikernels are great BUT: Unikernels lack full Linux Support App ● Hermitux: supports only 97 system calls Kernel LibOS + App ● OSv: ○ Fork() , execve() are not supported Hypervisor Hypervisor ○ Special files are not supported such as /proc ○ Signal mechanism is not complete ● Small kernel size ● Rumprun: only 37 curated applications ● Heavy ● Fast boot time ● Community is too small to keep it rolling ● Inefficient ● Improved performance ● Better security 2 Can Linux behave like a unikernel? 3 Lupine Linux 4 Lupine Linux ● Kernel mode Linux (KML) ○ Enables normal user process to run in kernel mode ○ Processes can still use system services such as paging and scheduling ○ App calls kernel routines directly without privilege transition costs ● Minimal patch to libc ○ Replace syscall instruction to call ○ The address of the called function is exported by the patched KML kernel using the vsyscall ○ No application changes/recompilation required 5 Boot time Evaluation Metrics Image size Based on: Unikernel benefits Memory footprint Application performance Syscall overhead 6 Configuration diversity ● 20 top apps on Docker hub (83% of all downloads) ● Only 19 configuration options required to run all 20 applications: lupine-general 7 Evaluation - Comparison configurations Lupine Cloud Operating Systems [Lupine-base + app-specific options] OSv general Linux-based Unikernels Kernel for 20 apps
    [Show full text]
  • Governance in Collaborative Open Source Software Development Organizations: a Comparative Analysis of Two Case Studies
    Governance in Collaborative Open Source Software Development Organizations: A Comparative Analysis of two Case Studies Master’s thesis Faculty of Business, Economics and Social Sciences University of Bern submitted to Dr. Matthias Stürmer Research Center for Digital Sustainability Institute of Information Systems by Winkelmann, Rahel from Siselen 11th semester Matriculation nr.: 09-127-788 Study address: Huberstrasse 22 3008 Bern (Tel. 078 758 58 96) (e-Mail: [email protected]) Bern, 20.01.2015 Abstract While loose cooperation among several actors is common in the open source sector, companies merging into a professionally governed collaborative open source software development organization across industries is an emerging phenomenon. The purpose of this thesis is to shed light on this new approach of software development by creating a framework for building a collaborative open source software development organization. A comparative analysis examines the governance models of two different collaborative open source software development organizations from the organizational, financial and legal perspective and reveals the autonomous and the affiliated organization type and their key characteristics. Based on these findings and by means of four expert interviews a framework consisting of eight criteria that need to be considered in order to build a collaborative open source software development organization is created. Zusammenfassung In der Open Source Branche ist es gängig, dass sich verschiedene Akteure zur Softwareentwicklung zu losen Konsortien zusammenschliessen. Unternehmen, welche sich im professionellen Rahmen zu einer Organisation zusammenschliessen um gemeinsam Open Source Software zu entwickeln, sind jedoch ein neues Phänomen. Der Zweck dieser Arbeit ist es Aufschluss über diesen neuen Ansatz von Softwareentwicklung zu geben.
    [Show full text]
  • Gvisor Is a Project to Restrict the Number of Syscalls That the Kernel and User Space Need to Communicate
    SED 820 Transcript EPISODE 820 [INTRODUCTION] [0:00:00.3] JM: The Linux operating system includes user space and kernel space. In user space, the user can create and interact with a variety of applications directly. In kernel space, the Linux kernel provides a stable environment in which device drivers interact with hardware and manage low-level resources. A Linux container is a virtualized environment that runs within user space. To perform an operation, a process in a container in user space makes a syscall, which is a system call into kernel space. This allows the container to have access to resources like memory and disk. Kernel space must be kept secure to ensure the operating system’s integrity. Linux includes hundreds of syscalls. Each syscall represents an interface between the user space and the kernel space. Security vulnerabilities can emerge from this wide attack surface of different syscalls and most applications only need a small number of syscalls to provide their required functionality. gVisor is a project to restrict the number of syscalls that the kernel and user space need to communicate. gVisor is a runtime layer between the user space container and the kernel space. gVisor reduces the number of syscalls that can be made into kernel space. The security properties of gVisor make it an exciting project today, but it is the portability features of gVisor that hint at a huge future opportunity. By inserting an interpreter interface between containers and the Linux kernel, gVisor presents the container world with an opportunity to run on operating systems other than Linux.
    [Show full text]
  • Surviving Software Dependencies
    practice DOI:10.1145/3347446 is additional code a programmer wants Article development led by queue.acm.org to call. Adding a dependency avoids repeating work: designing, testing, de- bugging, and maintaining a specific Software reuse is finally here unit of code. In this article, that unit of but comes with risks. code is referred to as a package; some systems use the terms library and mod- BY RUSS COX ule instead. Taking on externally written depen- dencies is not new. Most programmers have at one point in their careers had to go through the steps of manually Surviving installing a required library, such as C’s PCRE or zlib; C++’s Boost or Qt; or Java’s JodaTime or JUnit. These pack- ages contain high-quality, debugged Software code that required significant exper- tise to develop. For a program that needs the functionality provided by one of these packages, the tedious work of manually downloading, in- Dependencies stalling, and updating the package is easier than the work of redeveloping that functionality from scratch. The high fixed costs of reuse, however, mean manually reused packages tend to be big; a tiny package would be easier to reimplement. FOR DECADES, DISCUSSION of software reuse was more A dependency manager (a.k.a. pack- common than actual software reuse. Today, the situation age manager) automates the download- ing and installation of dependency is reversed: developers reuse software written by others packages. As dependency managers every day, in the form of software dependencies, and the make individual packages easier to download and install, the lower fixed situation goes mostly unexamined.
    [Show full text]
  • Administration Guide Administration Guide SUSE Linux Enterprise High Availability Extension 15 SP1 by Tanja Roth and Thomas Schraitle
    SUSE Linux Enterprise High Availability Extension 15 SP1 Administration Guide Administration Guide SUSE Linux Enterprise High Availability Extension 15 SP1 by Tanja Roth and Thomas Schraitle This guide is intended for administrators who need to set up, congure, and maintain clusters with SUSE® Linux Enterprise High Availability Extension. For quick and ecient conguration and administration, the product includes both a graphical user interface and a command line interface (CLI). For performing key tasks, both approaches are covered in this guide. Thus, you can choose the appropriate tool that matches your needs. Publication Date: September 24, 2021 SUSE LLC 1800 South Novell Place Provo, UT 84606 USA https://documentation.suse.com Copyright © 2006–2021 SUSE LLC and contributors. All rights reserved. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or (at your option) version 1.3; with the Invariant Section being this copyright notice and license. A copy of the license version 1.2 is included in the section entitled “GNU Free Documentation License”. For SUSE trademarks, see http://www.suse.com/company/legal/ . All other third-party trademarks are the property of their respective owners. Trademark symbols (®, ™ etc.) denote trademarks of SUSE and its aliates. Asterisks (*) denote third-party trademarks. All information found in this book has been compiled with utmost attention to detail. However, this does not guarantee complete accuracy. Neither SUSE
    [Show full text]
  • Firecracker: Lightweight Virtualization for Serverless Applications
    Firecracker: Lightweight Virtualization for Serverless Applications Alexandru Agache, Marc Brooker, Andreea Florescu, Alexandra Iordache, Anthony Liguori, Rolf Neugebauer, Phil Piwonka, and Diana-Maria Popa, Amazon Web Services https://www.usenix.org/conference/nsdi20/presentation/agache This paper is included in the Proceedings of the 17th USENIX Symposium on Networked Systems Design and Implementation (NSDI ’20) February 25–27, 2020 • Santa Clara, CA, USA 978-1-939133-13-7 Open access to the Proceedings of the 17th USENIX Symposium on Networked Systems Design and Implementation (NSDI ’20) is sponsored by Firecracker: Lightweight Virtualization for Serverless Applications Alexandru Agache Marc Brooker Andreea Florescu Amazon Web Services Amazon Web Services Amazon Web Services Alexandra Iordache Anthony Liguori Rolf Neugebauer Amazon Web Services Amazon Web Services Amazon Web Services Phil Piwonka Diana-Maria Popa Amazon Web Services Amazon Web Services Abstract vantage over traditional server provisioning processes: mul- titenancy allows servers to be shared across a large num- Serverless containers and functions are widely used for de- ber of workloads, and the ability to provision new func- ploying and managing software in the cloud. Their popularity tions and containers in milliseconds allows capacity to be is due to reduced cost of operations, improved utilization of switched between workloads quickly as demand changes. hardware, and faster scaling than traditional deployment meth- Serverless is also attracting the attention of the research com- ods. The economics and scale of serverless applications de- munity [21,26,27,44,47], including work on scaling out video mand that workloads from multiple customers run on the same encoding [13], linear algebra [20, 53] and parallel compila- hardware with minimal overhead, while preserving strong se- tion [12].
    [Show full text]
  • Erlang on Physical Machine
    on $ whoami Name: Zvi Avraham E-mail: [email protected] /ˈkɒm. pɑː(ɹ)t. mɛntl̩. aɪˌzeɪ. ʃən/ Physicalization • The opposite of Virtualization • dedicated machines • no virtualization overhead • no noisy neighbors – nobody stealing your CPU cycles, IOPS or bandwidth – your EC2 instance may have a Netflix “roommate” ;) • Mostly used by ARM-based public clouds • also called Bare Metal or HPC clouds Sandbox – a virtual container in which untrusted code can be safely run Sandbox examples: ZeroVM & AWS Lambda based on Google Native Client: A Sandbox for Portable, Untrusted x86 Native Code Compartmentalization in terms of Virtualization Physicalization No Virtualization Virtualization HW-level Virtualization Containerization OS-level Virtualization Sandboxing Userspace-level Virtualization* Cloud runs on virtual HW HARDWARE Does the OS on your Cloud instance still supports floppy drive? $ ls /dev on Ubuntu 14.04 AWS EC2 instance • 64 teletype devices? • Sound? • 32 serial ports? • VGA? “It’s DUPLICATED on so many LAYERS” Application + Configuration process* OS Middleware (Spring/OTP) Container Managed Runtime (JVM/BEAM) VM Guest Container OS Container Guest OS Hypervisor Hardware We run Single App per VM APPS We run in Single User mode USERS Minimalistic Linux OSes • Embedded Linux versions • DamnSmall Linux • Linux with BusyBox Min. Linux OSes for Containers JeOS – “Just Enough OS” • CoreOS • RancherOS • RedHat Project Atomic • VMware Photon • Intel Clear Linux • Hyper # of Processes and Threads per OS OSv + CLI RancherOS processes CoreOS threads
    [Show full text]
  • Architectural Implications of Function-As-A-Service Computing
    Architectural Implications of Function-as-a-Service Computing Mohammad Shahrad Jonathan Balkind David Wentzlaff Princeton University Princeton University Princeton University Princeton, USA Princeton, USA Princeton, USA [email protected] [email protected] [email protected] ABSTRACT Network Serverless computing is a rapidly growing cloud application model, popularized by Amazon’s Lambda platform. Serverless cloud ser- Scheduling vices provide fine-grained provisioning of resources, which scale Platform (priorwork) automatically with user demand. Function-as-a-Service (FaaS) appli- Queueing Management cations follow this serverless model, with the developer providing 35% decrease in IPC Interference their application as a set of functions which are executed in response due to interference 6x variation due to to a user- or system-generated event. Functions are designed to Memory BW invocation pattern 20x MPKI for be short-lived and execute inside containers or virtual machines, Branch MPKI >10x exec time short functions introducing a range of system-level overheads. This paper studies for short functions Cold Start Server the architectural implications of this emerging paradigm. Using (500ms cold start) Up to 20x (thispaper) Container the commercial-grade Apache OpenWhisk FaaS platform on real slowdown servers, this work investigates and identifies the architectural im- Native plications of FaaS serverless computing. The workloads, along with Execution Figure 1: We characterize the server-level overheads of the way that FaaS inherently interleaves short functions from many Function-as-a-Service applications, compared to native exe- tenants frustrates many of the locality-preserving architectural cution. This contrasts with prior work [2–5] which focused structures common in modern processors.
    [Show full text]
  • The Aurora Operating System
    The Aurora Operating System Revisiting the Single Level Store Emil Tsalapatis Ryan Hancock Tavian Barnes RCS Lab, University of Waterloo RCS Lab, University of Waterloo RCS Lab, University of Waterloo [email protected] [email protected] [email protected] Ali José Mashtizadeh RCS Lab, University of Waterloo [email protected] ABSTRACT KEYWORDS Applications on modern operating systems manage their single level stores, transparent persistence, snapshots, check- ephemeral state in memory, and persistent state on disk. En- point/restore suring consistency between them is a source of significant developer effort, yet still a source of significant bugs inma- ACM Reference Format: ture applications. We present the Aurora single level store Emil Tsalapatis, Ryan Hancock, Tavian Barnes, and Ali José Mash- (SLS), an OS that simplifies persistence by automatically per- tizadeh. 2021. The Aurora Operating System: Revisiting the Single sisting all traditionally ephemeral application state. With Level Store. In Workshop on Hot Topics in Operating Systems (HotOS recent storage hardware like NVMe SSDs and NVDIMMs, ’21), June 1-June 3, 2021, Ann Arbor, MI, USA. ACM, New York, NY, Aurora is able to continuously checkpoint entire applications USA, 8 pages. https://doi.org/10.1145/3458336.3465285 with millisecond granularity. Aurora is the first full POSIX single level store to han- dle complex applications ranging from databases to web 1 INTRODUCTION browsers. Moreover, by providing new ways to interact with Single level storage (SLS) systems provide persistence of and manipulate application state, it enables applications to applications as an operating system service. Their advantage provide features that would otherwise be prohibitively dif- lies in removing the semantic gap between the in-memory ficult to implement.
    [Show full text]