DOCSLIB.ORG

Enhancing Quality of Service Metrics for High Fan-In Node.Js Applications by Optimising the Network Stack

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Enhancing Quality of Service Metrics for High Fan-In Node.Js Applications by Optimising the Network Stack DEGREE PROJECT, IN COMPUTER SCIENCE , SECOND LEVEL LAUSANNE, SWITZERLAND 2015 Enhancing Quality of Service Metrics for High Fan-In Node.js Applications by Optimising the Network Stack LEVERAGING IX: THE DATAPLANE OPERATING SYSTEM FREDRIK PETER LILKAER KTH ROYAL INSTITUTE OF TECHNOLOGY SCHOOL OF COMPUTER SCIENCE AND COMMUNICATION (CSC) Enhancing Quality of Service Metrics for High Fan-in Node.js Applications by Optimising the Network Stack -Leveraging IX: The Dataplane Operating System FREDRIK PETER LILKAER DD221X, Master’s Thesis in Computer Science (30 ECTS credits) Degree Progr. in Computer Science and Engineering 300 credits Master Programme in Computer Science 120 credits Royal Institute of Technology year 2015 Supervisor at EPFL was Edouard Bugnion Supervisor at CSC wa s Carl-Henrik Ek Examiner wa s Johan Håstad Presented: 2015-10-01 Royal Institute of Technology School of Computer Science and Communication KTH CSC SE-100 44 Stockholm, Sweden URL: www.kth.se/csc Abstract This thesis investigates the feasibility of porting Node.js, a JavaScript web application framework and server, to IX, a data- plane operating system specifically developed to meet the needs of high performance microsecond-computing type of applications in a datacentre setting. We show that porting requires exten- sions to the IX kernel to support UDS polling, which we imple- ment. We develop a distributed load generator to benchmark the framework. The results show that running Node.js on IX improves throughput by up to 20.6%, latency by up to 5.23×, and tail latency by up to 5.68× compared to a Linux baseline. We show how server side request level reordering affect the la- tency distribution, predominantly in cases where the server is load saturated. Finally, due to various limitations of IX1, we are unable at this time to recommend running Node.js on IX in a production environment, despite improved metrics in all test cases. However, the limitations are not fundamental, and could be resolved in future work. Referat Förbättran av Quality of Service för högbelastade Node.js- webbapplikationer genom effektivare operativsystem Detta exjobb undersöker möjligheterna till att använda IX, ett specialiserat dataplansoperativsystem avsett för högpresterande datacentertillämpningar, för att köra Node.js, ett webapplika- tionramverk för JavaScript-applikationer. För att porta Node.js till IX krävs att vi utvidgar IX med funktionalitet för samtidig pollning av Unix Domain Sockets och nätverksflöden, vilket visas samt genomförs. Vidare utvecklas en distribuerad lastgenerator för att utvärdera applikationsramverket under IX jämfört baslin- je som utgörs av en omodifierad Linuxdistribution. Resultaten vi- sar att throughput förbättras med upp till 20.6%, latens upp till 5.23× och tail latency upp till 5.68×. Sedermera undersöker vi huruvida latensvariansen ökat på grund av request-omordningar på serversidan, vilket tycks vara fallet vid hög serverbelastning, även om andra faktorer tycks ha större inverkan vid låg server- belastning. Slutligen, även om alla storheter förbättrats vid alla observerade mätpunkter, kan ännu inte vidspredd adoption av IX för att köra Node.js applikationer rekommenderas, främst på grund av problem med horisontal skalning samt problem att ingå som frontend-server i en klassisk tiered-datacentre arkitektur. 1Mainly lack of outgoing TCP connections and multi-process execution, respectively preventing Node.js from acting as a frontend in a multi-tiered architecture and scaling horizontally within a single node. Acknowledgments Writing a thesis can be a long, and at times straining task. I would therefore like to thank the people that helped me achieve my thesis. First, I would like to thank the Data Center Systems laboratory at École Poly- technique Fédérale de Lausanne, EPFL, that allowed me to work with them for the duration of my thesis. In particular, I would like to thank my supervisor Edouard Bugnion, who offered invaluable advice every time I was stuck in my work. I would also like to thank Mia Primorac and George Prekas who I had the pleasure of working alongside, and who also withstood all my questions on IX. I would like to thank my supervisor at KTH, Carl-Henrik Ek, for offering good academic guidance and writing advice. Finally, I would like to thank all my friends of Lausanne for support and moti- vation during the semester. An extra thanks goes out to those of you that helped me to proofread. Contents Contents Glossary 1 Introduction 1 1.1 Problem Statement . 2 1.2 Contribution . 3 2 Background 5 2.1 Operating Systems . 5 2.2 The IX Dataplane Operating System . 6 2.2.1 Requirements and Motivations . 7 2.2.2 What is a Dataplane Operating System? . 8 2.2.3 Results . 8 2.3 Web Servers . 9 2.3.1 Apache, the Traditional Forking Web Server . 9 2.3.2 Nginx - the Event Driven Web Server . 9 2.3.3 Node.js . 10 2.4 Queueing Theory . 10 3 Software Foundation 13 3.1 The IX Dataplane Operating System . 13 3.1.1 Architectural Overview . 13 3.1.2 Dune Process Virtualisation . 14 3.1.3 Execution Model . 15 3.1.4 IX System Call API . 16 3.1.5 IX Event Conditions . 17 3.1.6 libix Userspace API . 17 3.1.7 Limitations . 18 3.2 Node.js . 18 3.2.1 V8 Javascript Engine . 19 3.2.2 libuv . 19 4 Design 25 CONTENTS 4.1 Design Overview . 25 4.2 Limitations . 25 4.3 Modifications of IX . 26 4.3.1 Motivation for IX Kernel Extensions . 26 4.3.2 Kernel Extension . 27 4.3.3 libix . 28 4.4 Modifications of Node.js . 28 4.4.1 Modifications of libuv . 28 4.4.2 Modifications of the V8 Javascript Engine . 34 5 Evaluation 35 5.1 Results . 35 5.1.1 Test Methodology . 35 5.1.2 Performance Metrics . 36 5.1.3 A Note on Poisson Distributed Arrival Rates . 37 5.1.4 Load Scaling . 37 5.1.5 Connection Scalability . 38 5.2 Result Tracing . 38 5.2.1 Throughput Increase . 39 5.2.2 Reordering & Tail Latency . 39 6 Discussion 43 6.1 Related Work . 45 6.2 Lessons Learned . 45 6.3 Future Work . 46 6.4 Conclusion . 47 Bibliography 49 A Resources 53 A.1 libuv - ix . 53 A.2 Node.js . 53 B dialog - high concurrency rate controlled poisson distributed load generator 55 B.1 Purpose . 55 B.2 Implementation . 55 B.3 Evaluation . 56 B.4 Resources . 58 Glossary API Application Programming Interface. 2, 16–20, 24, 25, 28, 29, 31, 47 ASLR Address Space Layout Randomisation. 34, 35, 53 FIFO First-In, First-Out. 11, 41 HTTP HyperText Transfer Protocol. 18, 35, 56 IPC Inter-Process Communication. 5, 46 libOS library Operating System. 6, 15, 47 LIFO Last-In, First-Out. 12 NIC Network Interface Controller. 26, 44 OS Operating System. 5, 19, 43 RPC Remote Procedure Call. 16, 28 RSS Receive Side Scaling. 15 SIRO Service in Random Order. 12 SLA Service Level Agreement. 3, 36, 38 TCP Transmission Control Protocol. 20, 22, 29, 32, 43 TLB Translation Lookaside Buffer. 14 UDP User Datagram Protocol. 43 UDS Unix Domain Socket. 20, 24, 26–28, 34, 47 Chapter 1 Introduction Almost everyone have probably heard about Moore’s law in one form or another; that computers double in processing power approximately every 18 months1. Con- sequently we should, by now, be free of performance problems since our computers ought to be super fast, given an exponential growth in processing power. And they are. The problem is just that we are constantly telling our computers to solve big- ger, and/or harder problems. Around the year 2004, it stopped to be efficient to scale CPU processing performance vertically, that is increasing the clock frequency. As a result, we are now constructing software to make use of multi-core processors, and we are engineering large, complex, distributed systems to deal with the gigantic datasets that we like to call “big data”. We find that it is important to bound the end-to-end latency, particularly in such systems. End-to-end latency is a key per- formance indicator and has a direct correlation with user experience and thus, for a commercial system, both customer conversion and customer retention, in particular in a realtime/online system In such distributed systems, computation is divided between multiple entities, which may be spread across a pleathoria of machines within a single - or across - datacentre(s). Therefore, one way to minimise the end-to-end latency and to control its distribution is to attempt to bound the latency of every participating component. The motivation is that latency and variance in latency is induced in every step of communication along the execution path. Furthermore, in current computer cluster deployments, energy accounts for a significant portion of operational expenses. Consequently, if we can engineer systems that are able to perform the required tasks more efficiently, they can run with fewer hardware resources and thus consume less energy resources. Therefore, it is still desired to improve the efficiency of our systems, even if we have at our disposal, extremely powerful computational resources. In this work we explore a method to improve the performance of web servers based on the Node.js application framework, that may or may not, be used in such a distributed setting as described in the first paragraph. The performance met- 1 The number of transistors on a die doubles approximately every 18 months. 1 CHAPTER 1. INTRODUCTION rics/Quality of Service metrics we study are mainly latency and its distribution as motivated in the second paragraph, and throughput. Throughput is the num- ber of transactions per time unit, and exhibits correlation with energy efficiency requirements as described in the third paragraph.
Load more

Recommended publications
  • A Practical UNIX Capability System
    A Practical UNIX Capability System Adam Langley <[email protected]> 22nd June 2005 ii Abstract This report seeks to document the development of a capability security system based on a Linux kernel and to follow through the implications of such a system. After defining terms, several other capability systems are discussed and found to be excellent, but to have too high a barrier to entry. This motivates the development of the above system. The capability system decomposes traditionally monolithic applications into a number of communicating actors, each of which is a separate process. Actors may only communicate using the capabilities given to them and so the impact of a vulnerability in a given actor can be reasoned about. This design pattern is demonstrated to be advantageous in terms of security, comprehensibility and mod- ularity and with an acceptable performance penality. From this, following through a few of the further avenues which present themselves is the two hours traffic of our stage. Acknowledgments I would like to thank my supervisor, Dr Kelly, for all the time he has put into cajoling and persuading me that the rest of the world might have a trick or two worth learning. Also, I’d like to thank Bryce Wilcox-O’Hearn for introducing me to capabilities many years ago. Contents 1 Introduction 1 2 Terms 3 2.1 POSIX ‘Capabilities’ . 3 2.2 Password Capabilities . 4 3 Motivations 7 3.1 Ambient Authority . 7 3.2 Confused Deputy . 8 3.3 Pervasive Testing . 8 3.4 Clear Auditing of Vulnerabilities . 9 3.5 Easy Configurability .
    [Show full text]
  • Knot DNS Resolver Release 1.2.0
    Knot DNS Resolver Release 1.2.0 CZ.NIC Labs Apr 25, 2017 Contents 1 Building project 3 1.1 Installing from packages.........................................3 1.2 Platform considerations.........................................3 1.3 Requirements...............................................3 1.4 Building from sources..........................................5 1.5 Getting Docker image..........................................7 2 Knot DNS Resolver library 9 2.1 Requirements...............................................9 2.2 For users.................................................9 2.3 For developers..............................................9 2.4 Writing layers.............................................. 11 2.5 APIs in Lua................................................ 12 2.6 API reference............................................... 15 3 Knot DNS Resolver daemon 47 3.1 Enabling DNSSEC............................................ 47 3.2 CLI interface............................................... 48 3.3 Scaling out................................................ 48 3.4 Running supervised........................................... 49 3.5 Configuration............................................... 49 3.6 Using CLI tools............................................. 64 4 Knot DNS Resolver modules 67 4.1 Static hints................................................ 67 4.2 Statistics collector............................................ 69 4.3 Query policies.............................................. 71 4.4 Views and ACLs............................................
    [Show full text]
  • Mysql NDB Cluster 7.5.16 (And Later)
    Licensing Information User Manual MySQL NDB Cluster 7.5.16 (and later) Table of Contents Licensing Information .......................................................................................................................... 2 Licenses for Third-Party Components .................................................................................................. 3 ANTLR 3 .................................................................................................................................... 3 argparse .................................................................................................................................... 4 AWS SDK for C++ ..................................................................................................................... 5 Boost Library ............................................................................................................................ 10 Corosync .................................................................................................................................. 11 Cyrus SASL ............................................................................................................................. 11 dtoa.c ....................................................................................................................................... 12 Editline Library (libedit) ............................................................................................................. 12 Facebook Fast Checksum Patch ..............................................................................................
    [Show full text]
  • Ultimate++ Forum Probably with These Two Variables You Could Try to Integrate D-BUS
    Subject: DBus integration -- need help Posted by jlfranks on Thu, 20 Jul 2017 15:30:07 GMT View Forum Message <> Reply to Message We are trying to add a DBus server to existing large U++ application that runs only on Linux. I've converted from X11 to GTK for Upp project config to be compatible with GIO dbus library. I've created a separate thread for the DBus server and ran into problems with event loop. I've gutted my DBus server code of everything except what is causing an issue. DBus GIO examples use GMainLoop in order for DBus to service asynchronous events. Everything compiles and runs except the main UI is not longer visible. There must be a GTK main loop already running and I've stepped on it with this code. Is there a way for me to obtain a pointer to the UI main loop and use it with my DBus server? How/where can I do that? Code snipped example as follows: ---- code snippet ---- myDBusServerMainThread() { //========================================================== ===== // // Enter main service loop for this thread // while (not needsExit ()) { // colaborate with join() GMainLoop *loop; loop = g_main_loop_new(NULL, FALSE); g_main_loop_run(loop); } } Subject: Re: DBus integration -- need help Posted by Klugier on Thu, 20 Jul 2017 22:30:17 GMT View Forum Message <> Reply to Message Hello, You could try use following methods of Ctrl to obtain gtk & gdk handlers: GdkWindow *gdk() const { return top ? top->window->window : NULL; } GtkWindow *gtk() const { return top ? (GtkWindow *)top->window : NULL; } Page 1 of 3 ---- Generated from Ultimate++ forum Probably with these two variables you could try to integrate D-BUS.
    [Show full text]
  • Fundamentals of Xlib Programming by Examples
    Fundamentals of Xlib Programming by Examples by Ross Maloney Contents 1 Introduction 1 1.1 Critic of the available literature . 1 1.2 The Place of the X Protocol . 1 1.3 X Window Programming gotchas . 2 2 Getting started 4 2.1 Basic Xlib programming steps . 5 2.2 Creating a single window . 5 2.2.1 Open connection to the server . 6 2.2.2 Top-level window . 7 2.2.3 Exercises . 10 2.3 Smallest Xlib program to produce a window . 10 2.3.1 Exercises . 10 2.4 A simple but useful X Window program . 11 2.4.1 Exercises . 12 2.5 A moving window . 12 2.5.1 Exercises . 15 2.6 Parts of windows can disappear from view . 16 2.6.1 Testing overlay services available from an X server . 17 2.6.2 Consequences of no server overlay services . 17 2.6.3 Exercises . 23 2.7 Changing a window’s properties . 23 2.8 Content summary . 25 3 Windows and events produce menus 26 3.1 Colour . 26 3.1.1 Exercises . 27 i CONTENTS 3.2 A button to click . 29 3.3 Events . 33 3.3.1 Exercises . 37 3.4 Menus . 37 3.4.1 Text labelled menu buttons . 38 3.4.2 Exercises . 43 3.5 Some events of the mouse . 44 3.6 A mouse behaviour application . 55 3.6.1 Exercises . 58 3.7 Implementing hierarchical menus . 58 3.7.1 Exercises . 67 3.8 Content summary . 67 4 Pixmaps 68 4.1 The pixmap resource .
    [Show full text]
  • Safeguard for Privileged Passwords 6.0.9 LTS Release Notes
    Safeguard for Privileged Passwords 6.0.9 LTS Release Notes 03 March 2021, 06:20 These release notes provide information about the Safeguard for Privileged Passwords 6.0.9 LTS release. If you are updating a Safeguard for Privileged Passwords version prior to this release, read the release notes for the version found at: One Identity Safeguard for Privileged Passwords Technical Documentation. For the most recent documents and product information, see One Identity Safeguard for Privileged Passwords Technical Documentation. Release options Safeguard for Privileged Passwords includes two release versions: l Long Term Support (LTS) release, version 6.0.9 LTS l Feature release, version 6.9 The versions align with Safeguard for Privileged Sessions. For more information, see Long Term Support (LTS) and Feature Releases on page 13. About this release Safeguard for Privileged Passwords Version 6.0.9 LTS is a minor LTS release with resolved issues. For more details on the features and resolved issues, see: Safeguard for Privileged Passwords 6.0.9 LTS 1 Release Notes l Resolved issues NOTE: For a full list of key features in Safeguard for Privileged Passwords, see the Safeguard for Privileged Passwords Administration Guide. About the Safeguard product line The Safeguard for Privileged Passwords Appliance is built specifically for use only with the Safeguard for Privileged Passwords privileged management software, which is pre- installed and ready for immediate use. The appliance is hardened to ensure the system is secured at the hardware, operating system, and software levels. The hardened appliance approach protects the privileged management software from attacks while simplifying deployment and ongoing management and shortening the time frame to value.
    [Show full text]
  • Efficient Parallel I/O on Multi-Core Architectures
    Lecture series title/ lecture title Efficient parallel I/O on multi-core architectures Adrien Devresse CERN IT-SDC-ID Thematic CERN School of Computing 2014 1 Author(s) names – Affiliation Lecture series title/ lecture title How to make I/O bound application scale with multi-core ? What is an IO bound application ? → A server application → A job that accesses big number of files → An application that uses intensively network 2 Author(s) names – Affiliation Lecture series title/ lecture title Stupid example: Simple server monothreaded // create socket socket_desc = socket(AF_INET , SOCK_STREAM , 0); // bind the socket bind(socket_desc,(struct sockaddr *)&server , sizeof(server)); listen(socket_desc , 100); //accept connection from an incoming client while(1){ // declarations client_sock = accept(socket_desc, (struct sockaddr *)&client, &c); //Receive a message from client while( (read_size = recv(client_sock , client_message , 2000 , 0)) > 0{ // Wonderful, we have a client, do some useful work std::string msg("hello bob"); write(client_sock, msg.c_str(), msg.size()); } } 3 Author(s) names – Affiliation Lecture series title/ lecture title Stupid example: Let's make it parallel ! int main(int argc, char** argv){ // creat socket void do_work(int socket){ socket_desc = socket(AF_INET , SOCK_STREAM , 0); //Receive a message while( (read_size = // bind the socket recv(client_sock , bind(socket_desc, server , sizeof(server)); client_message , 2000 , 0)) > 0{ listen(socket_desc , 100); // Wonderful, we have a client // useful works //accept connection
    [Show full text]
  • Message Passing and Network Programming
    Message Passing and Network Programming Advanced Operating Systems Lecture 13 Colin Perkins | https://csperkins.org/ | Copyright © 2017 | This work is licensed under the Creative Commons Attribution-NoDerivatives 4.0 International License. To view a copy of this license, visit http://creativecommons.org/licenses/by-nd/4.0/ or send a letter to Creative Commons, PO Box 1866, Mountain View, CA 94042, USA. Lecture Outline • Actors, sockets, and network protocols • Asynchronous I/O frameworks • Higher level abstractions Colin Perkins | https://csperkins.org/ | Copyright © 2017 2 Message Passing and Network Protocols • Recap: • Actor-based framework for message passing Send to • Each actor has a receive loop other actors Mailbox Actor Calls to one function per state Queue • Receive Message • Messages delivered by runtime system; Receiver processed sequentially Message Done Message Process • Actor can send messages in reply; Message Dispatcher return identity of next state Dequeue • Can we write network code this way? Request next • Send data by sending a message to an actor representing a socket • Receive messages representing data received on a socket Colin Perkins | https://csperkins.org/ | Copyright © 2017 3 Integrating Actors and Sockets Sending Thread Send to other actors Encoder Network Socket Mailbox Actor Queue Parser Receive Message Receiver Message Done Receiving Thread Message Process Message Dispatcher • Conceptually straightforward to integrate Dequeue actors with network code Request next • Runtime system maintains sending and
    [Show full text]
  • Copyright by Tongliang Liao 2017
    Copyright by Tongliang Liao 2017 The Thesis committee for Tongliang Liao certifies that this is the approved version of the following thesis: TAI: Threaded Asynchronous I/O Library for Performance and Portability APPROVED BY SUPERVISING COMMITTEE: Vijaychidambaram Velayudhan Pillai, Supervisor Simon Peter TAI: Threaded Asynchronous I/O Library for Performance and Portability by Tongliang Liao Thesis Presented to the Faculty of the Graduate School of the University of Texas at Austin in Partial Fulfillment of the Requirements for the Degree of Master of Science in Computer Science The University of Texas at Austin Dec 2017 TAI: Threaded Asynchronous I/O Library for Performance and Portability by Tongliang Liao, M.S.C.S The University of Texas at Austin, 2017 Supervisor: Vijaychidambaram Velayudhan Pillai In this paper, we investigate the behavior and performance of disk I/O using different types of libraries. We analyze the scenario where we can benefit from asyn- chronous I/O, and propose our cross-platform library design called TAI (Threaded Async I/O). TAI is designed to be a C++17 library with developer-friendly API. Our benchmark shows it can out-perform other libraries when asynchronous I/O is beneficial, and keep competitive speed in other cases. It also demonstrates TAI’s ability to retrieve 20% - 60% speedup on poorly scaled serial code by a simple library replacement. iv Table of Contents 1 Introduction 1 1.1 Related Work .................................................................................. 2 1.2 Background ..................................................................................... 2 1.2.1 POSIX Sync I/O ................................................................... 3 1.2.2 POSIX AIO .......................................................................... 3 1.2.3 C/C++ Standard I/O Functions............................................
    [Show full text]
  • A Sense of Time for Node.Js: Timeouts As a Cure for Event Handler Poisoning
    A Sense of Time for Node.js: Timeouts as a Cure for Event Handler Poisoning Anonymous Abstract—The software development community has begun to new Denial of Service attack that can be used against EDA- adopt the Event-Driven Architecture (EDA) to provide scalable based services. Our Event Handler Poisoning attack exploits web services. Though the Event-Driven Architecture can offer the most important limited resource in the EDA: the Event better scalability than the One Thread Per Client Architecture, Handlers themselves. its use exposes service providers to a Denial of Service attack that we call Event Handler Poisoning (EHP). The source of the EDA’s scalability is also its Achilles’ heel. Multiplexing unrelated work onto the same thread re- This work is the first to define EHP attacks. After examining EHP vulnerabilities in the popular Node.js EDA framework and duces overhead, but it also moves the burden of time sharing open-source npm modules, we explore various solutions to EHP- out of the thread library or operating system and into the safety. For a practical defense against EHP attacks, we propose application itself. Where OTPCA-based services can rely on Node.cure, which defends a large class of Node.js applications preemptive multitasking to ensure that resources are shared against all known EHP attacks by making timeouts a first-class fairly, using the EDA requires the service to enforce its own member of the JavaScript language and the Node.js framework. cooperative multitasking [89]. An EHP attack identifies a way to defeat the cooperative multitasking used by an EDA-based Our evaluation shows that Node.cure is effective, broadly applicable, and offers strong security guarantees.
    [Show full text]
  • Introduction to Asynchronous Programming
    Introduction to Asynchronous Programming In this document we introduce an asynchronous model for concurrent programming. For certain appli- cations, an asynchronous model may yield performance benefits over traditional multithreading. Much of the material presented in this document is taken from Dave Peticola’s excellent introduction to Twisted1, a Python framework for asynchronous programming. 1 The Models We will start by reviewing two (hopefully) familiar models in order to contrast them with the asynchronous model. By way of illustration we will imagine a program that consists of three conceptually distinct tasks which must be performed to complete the program. Note I am using task in the non-technical sense of something that needs to be done. The first model we will look at is the single-threaded synchronous model, in Figure 1 below: Figure 1: The single-threaded synchronous model This is the simplest style of programming. Each task is performed one at a time, with one finishing completely before another is started. And if the tasks are always performed in a definite order, the imple- mentation of a later task can assume that all earlier tasks have finished without errors, with all their output available for use — a definite simplification in logic. We can contrast the single-threaded synchronous model with the multi-threaded synchronous model illustrated in Figure 2. In this model, each task is performed in a separate thread of control. The threads are managed by the operating system and may, on a system with multiple processors or multiple cores, run truly concurrently, 1http://krondo.com/?page_id=1327 1 CS168 Async Programming Figure 2: The threaded model or may be interleaved together on a single processor.
    [Show full text]
  • Memc3: Compact and Concurrent Memcache with Dumber Caching and Smarter Hashing
    MemC3: Compact and Concurrent MemCache with Dumber Caching and Smarter Hashing Bin Fan, David G. Andersen, Michael Kaminsky∗ Carnegie Mellon University, ∗Intel Labs Abstract Standard Memcached, at its core, uses a typical hash table design, with linked-list-based chaining to handle This paper presents a set of architecturally and workload- collisions. Its cache replacement algorithm is strict LRU, inspired algorithmic and engineering improvements also based on linked lists. This design relies on locking to the popular Memcached system that substantially to ensure consistency among multiple threads, and leads improve both its memory efficiency and throughput. to poor scalability on multi-core CPUs [11]. These techniques—optimistic cuckoo hashing, a com- This paper presents MemC3 (Memcached with pact LRU-approximating eviction algorithm based upon CLOCK and Concurrent Cuckoo Hashing), a complete CLOCK, and comprehensive implementation of opti- redesign of the Memcached internals. This re-design mistic locking—enable the resulting system to use 30% is informed by and takes advantage of several observa- less memory for small key-value pairs, and serve up to tions. First, architectural features can hide memory access 3x as many queries per second over the network. We latencies and provide performance improvements. In par- have implemented these modifications in a system we ticular, our new hash table design exploits CPU cache call MemC3—Memcached with CLOCK and Concur- locality to minimize the number of memory fetches re- rent Cuckoo hashing—but believe that they also apply quired to complete any given operation; and it exploits more generally to many of today’s read-intensive, highly instruction-level and memory-level parallelism to overlap concurrent networked storage and caching systems.
    [Show full text]