scalability in the clouds a myth or reality
play

Scalability in the Clouds! A Myth or Reality? Sanidhya Kashyap , - PowerPoint PPT Presentation

May 20, 2023 •290 likes •1.21k views

Scalability in the Clouds! A Myth or Reality? Sanidhya Kashyap , Changwoo Min, Taesoo Kim Programmer's Paradise? A programmer day-to-day task: program compilation, like Linux kernel compilation. Relies on Buildbot to complete the job ASAP!

  1. Scalability in the Clouds! A Myth or Reality? Sanidhya Kashyap , Changwoo Min, Taesoo Kim

  2. Programmer's Paradise? ● A programmer day-to-day task: program compilation, like Linux kernel compilation. ● Relies on Buildbot to complete the job ASAP! ● Expects the job to complete sooner with increasing core count. – With respect to vertical scalability, a parallel job with no sequential bottleneck should scale with increasing core count.

  3. Programmer's Paradise? ● A programmer day-to-day task: program compilation, like Linux kernel compilation. ● Relies on Buildbot to complete the job ASAP! How about using Cloud providers for our fun ● Expects the job to complete sooner with and their profjt? increasing core count. – With respect to vertical scalability, a parallel job with no sequential bottleneck should scale with increasing core count.

  4. Clouds Trend ● Trend is changing Larger instances (40 → vCPUs) are available. ● Will Buildbot really scale? 50 vCPUs 40 40 32 32 32 32 30 20 16 8 8 10 4 1 0 2007 2009 2011 2013 2015 2006 2008 2010 2012 2014

  5. Scalability Behavior in the Clouds 140 120 EC2 100 builds / hour 80 60 40 20 0 4 8 12 16 20 24 28 32 #vCPUs

  6. Scalability Behavior in the Clouds 140 120 EC2 100 GCE builds / hour 80 60 40 20 0 4 8 12 16 20 24 28 32 #vCPUs

  7. Scalability Behavior in the Clouds 140 EC2 120 GCE 100 builds / hour Azure 80 60 40 20 0 4 8 12 16 20 24 28 32 #vCPUs

  8. Scalability Behavior in the Clouds 140 EC2 120 GCE 100 Azure builds / hour 16-core E5 80 60 40 20 0 4 8 12 16 20 24 28 32 #vCPUs

  9. Scalability Behavior in the Clouds 140 EC2 120 GCE 100 Azure builds / hour 16-core E5 80 60 40 20 0 4 8 12 16 20 24 28 32 #vCPUs

  10. Scalability Behavior in the Clouds 140 EC2 120 GCE 100 Azure builds / hour 16-core E5 80 60 40 20 0 4 8 12 16 20 24 28 32 #vCPUs

  11. Scalability Behavior in the Clouds 140 EC2 120 GCE 100 Azure builds / hour 16-core E5 80 60 40 20 0 4 8 12 16 20 24 28 32 #vCPUs

  12. Scalability Behavior in the Clouds 140 EC2 ? 120 GCE 100 Azure builds / hour 16-core E5 80 60 40 20 0 4 8 12 16 20 24 28 32 #vCPUs

  13. Scalability Behavior in VMs with Higher-core count 250 Host 200 builds / hour 150 100 50 0 20 40 60 80 100 120 140 160 #vCPUs

  14. Scalability Behavior in VMs with Higher-core count 250 Host Guest 200 builds / hour 150 100 50 0 20 40 60 80 100 120 140 160 #vCPUs

  15. Scalability Behavior in VMs with Higher-core count 250 Host Guest 200 builds / hour 150 100 6.7x 50 0 20 40 60 80 100 120 140 160 #vCPUs

  16. Why? 200 ● Performance 150 builds / hour degradation occurs 100 Guest due to drastic 50 increase in VMEXITS 0 ( halt exits ). 0 20 40 60 80 100 120 140 160 1400 #halt exits x 1000 1200 1000 800 600 Guest 400 200 0 20 40 60 80 100 120 140 160 #vCPUs

  17. Why? 200 ● Performance 150 builds / hour degradation occurs 100 Guest due to drastic 50 increase in VMEXITS 0 ( halt exits ). 0 20 40 60 80 100 120 140 160 1400 #halt exits x 1000 1200 1000 800 600 Spinlock is sleeping! Guest 400 200 0 20 40 60 80 100 120 140 160 #vCPUs

  18. Spinlock Evolution in the Linux Kernel

  19. Spinlock Evolution in the Linux Kernel Test-and- Test-And-Set spinlock

  20. Spinlock Evolution in the Linux Kernel Test-and- 2.6.25 (April Test-And-Set Fairness 2008) spinlock Ticket spinlock

  21. Spinlock Evolution in the Linux Kernel Test-and- 2.6.25 (April 3.15 (July 2014) Test-And-Set Fairness 2008) Shared cacheline qspinlock, variant spinlock Ticket spinlock of MCS lock (yet to contention be merged)

  22. Spinlock Evolution in the Linux Kernel 3.11 ( 2013) Paravirtual Ticket spinlock Test-and- 2.6.25 (April 3.15 (July 2014) Test-And-Set Fairness 2008) Shared cacheline qspinlock, variant spinlock Ticket spinlock of MCS lock (yet to contention be merged)

  23. Spinlock Evolution in the Linux Kernel 3.11 ( 2013) 4.0 (May, 2015) Paravirtual Ticket OTicket spinlock Test-and- 2.6.25 (April 3.15 (July 2014) Test-And-Set Fairness 2008) Shared cacheline qspinlock, variant spinlock Ticket spinlock of MCS lock (yet to contention be merged)

  24. Ticket Spinlock #defj fjne SPIN_THRESHOLD (1 << 15) int head = 0; int tail = 0; F&I(*addr) { old = *addr; int threshold = SPIN_THRESHOLD; *addr++; void lock() { return old; my_ticket = F&I(tail); } for( ; ; ) { int count = threshold; do { head if(my_ticket == head); goto out; } while(--count); } out: ; tail } void unlock() { head++; } ● Guaranteed FIFO ordering. ● Mitigates starvation with increasing core count.

  25. Ticket Spinlock #defj fjne SPIN_THRESHOLD (1 << 15) int head = 0; int tail = 0; F&I(*addr) { old = *addr; int threshold = SPIN_THRESHOLD; *addr++; void lock() { return old; my_ticket = F&I(tail); } for( ; ; ) { int count = threshold; do { head if(my_ticket == head); goto out; } while(--count); } out: ; tail } void unlock() { head++; } ● Guaranteed FIFO ordering. ● Mitigates starvation with increasing core count.

  26. Ticket Spinlock #defj fjne SPIN_THRESHOLD (1 << 15) int head = 0; int tail = 0; F&I(*addr) { old = *addr; int threshold = SPIN_THRESHOLD; *addr++; void lock() { return old; my_ticket = F&I(tail); } for( ; ; ) { int count = threshold; do { head if(my_ticket == head); goto out; } while(--count); } out: ; tail } void unlock() { head++; } ● Guaranteed FIFO ordering. ● Mitigates starvation with increasing core count.

  27. Ticket Spinlock #defj fjne SPIN_THRESHOLD (1 << 15) int head = 0; int tail = 0; F&I(*addr) { old = *addr; int threshold = SPIN_THRESHOLD; *addr++; void lock() { return old; my_ticket = F&I(tail); } for( ; ; ) { int count = threshold; do { head if(my_ticket == head); goto out; } while(--count); } out: ; tail } void unlock() { head++; } ● Guaranteed FIFO ordering. ● Mitigates starvation with increasing core count.

  28. Ticket Spinlock #defj fjne SPIN_THRESHOLD (1 << 15) int head = 0; int tail = 0; F&I(*addr) { old = *addr; int threshold = SPIN_THRESHOLD; *addr++; void lock() { return old; my_ticket = F&I(tail); } for( ; ; ) { int count = threshold; do { head if(my_ticket == head); goto out; } while(--count); } out: ; tail } void unlock() { head++; } ● Guaranteed FIFO ordering. ● Mitigates starvation with increasing core count.

  29. Ticket Spinlock #defj fjne SPIN_THRESHOLD (1 << 15) int head = 0; int tail = 0; F&I(*addr) { old = *addr; int threshold = SPIN_THRESHOLD; *addr++; void lock() { return old; my_ticket = F&I(tail); } for( ; ; ) { int count = threshold; do { head if(my_ticket == head); goto out; } while(--count); } out: ; tail } void unlock() { head++; } ● Guaranteed FIFO ordering. ● Mitigates starvation with increasing core count.

  30. Ticket Spinlock #defj fjne SPIN_THRESHOLD (1 << 15) int head = 0; int tail = 0; F&I(*addr) { old = *addr; int threshold = SPIN_THRESHOLD; *addr++; void lock() { return old; my_ticket = F&I(tail); } for( ; ; ) { int count = threshold; do { head if(my_ticket == head); goto out; } while(--count); } out: ; tail } void unlock() { head++; } ● Guaranteed FIFO ordering. ● Mitigates starvation with increasing core count.

  31. Ticket Spinlock #defj fjne SPIN_THRESHOLD (1 << 15) int head = 0; int tail = 0; F&I(*addr) { old = *addr; int threshold = SPIN_THRESHOLD; *addr++; void lock() { return old; my_ticket = F&I(tail); } for( ; ; ) { int count = threshold; do { head if(my_ticket == head); goto out; } while(--count); } out: ; tail } void unlock() { head++; } ● Guaranteed FIFO ordering. ● Mitigates starvation with increasing core count.

  32. Complexity of Ticket Spinlock in Virtualized Environment ● vCPUs are scheduled by host Guest OS scheduler. ● Semantic gap between the vCPU 1 vCPU 2 hypervisor and guest OS. Hypervisor CPU 1

  33. Complexity of Ticket Spinlock in Virtualized Environment ● Lock Holder Preemption: vCPU holding the lock gets preempted.

  34. Complexity of Ticket Spinlock in Virtualized Environment ● Lock Holder Preemption: vCPU holding the lock gets preempted. Scheduled Preempted

  35. Complexity of Ticket Spinlock in Virtualized Environment ● Lock Holder Preemption: vCPU holding the lock gets preempted. 0 head = 0 tail = 1 Scheduled Preempted

  36. Complexity of Ticket Spinlock in Virtualized Environment ● Lock Holder Preemption: vCPU holding the lock gets preempted. 0 1 head = 0 tail = 2 Scheduled Preempted

  37. Complexity of Ticket Spinlock in Virtualized Environment ● Lock Holder Preemption: vCPU holding the lock gets preempted. 0 1 2 head = 0 Scheduled Preempted

  38. Complexity of Ticket Spinlock in Virtualized Environment ● Lock Holder Preemption: vCPU holding the lock gets preempted. 1 2 head = 1 Scheduled Preempted

  39. Complexity of Ticket Spinlock in Virtualized Environment ● Lock Holder Preemption: vCPU holding the lock gets preempted. 1 2 head = 1 Scheduled Preempted

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend

Gifted 101 SWEPP Presentation October 6, 2015 Gifted: Myth vs Reality Myth or Reality? 1.

Gifted 101 SWEPP Presentation October 6, 2015 Gifted: Myth vs Reality Myth or Reality? 1.

Gifted 101 SWEPP Presentation October 6, 2015 Gifted: Myth vs Reality Myth or Reality? 1. Gifted students will achieve without guidance. Myth: Consider a star athlete - would they train for the Olympics without a coach? Many gifted students

495 views • 21 slides
Through Silicon Via for 3D integra5on Myth or reality ?

Through Silicon Via for 3D integra5on Myth or reality ?

Through Silicon Via for 3D integra5on Myth or reality ? Advanced Seminar Computer Engineering WS 2012/2013 Solience

921 views • 35 slides
Scalability and Replication Marco Serafini COMPSCI 532 Lecture 13 Scalability 2 Scalability

Scalability and Replication Marco Serafini COMPSCI 532 Lecture 13 Scalability 2 Scalability

Scalability and Replication Marco Serafini COMPSCI 532 Lecture 13 Scalability 2 Scalability Ideal world Linear scalability Speedup Reality Ideal Bottlenecks For example: central coordinator When do we stop

938 views • 36 slides
Medieval Europe: Myth and Reality In film and in literature, medieval life seems heroic,

Medieval Europe: Myth and Reality In film and in literature, medieval life seems heroic,

Medieval Europe: Myth and Reality In film and in literature, medieval life seems heroic, entertaining, and romantic. We think of knights in shining armor, banquets, minstrels, kings, queens, and glorious pageantry. In reality, life in

341 views • 6 slides
European Cooperation in the Youth Field: Myth or Reality? Presentation Module 2 Inside the EU

European Cooperation in the Youth Field: Myth or Reality? Presentation Module 2 Inside the EU

YouthMetre - Training Materials European Cooperation in the Youth Field: Myth or Reality? Presentation Module 2 Inside the EU Youth Policy EU has full powers about youth policy MYTH EU does not have an exclusive competence about youth

364 views • 12 slides
Living to 100 a Myth or a Reality? Conference on retirement, institutional investment and

Living to 100 a Myth or a Reality? Conference on retirement, institutional investment and

Canada Pension Plan Enhancement Living to 100 a Myth or a Reality? Conference on retirement, institutional investment and personal finances by Jean-Claude Mnard, Chief Actuary, OCA, OSFI Presentation to the OSFI Actuarial Division by

298 views • 16 slides
When you look up into the sky, you will often see clouds. No two clouds are the same, and there

When you look up into the sky, you will often see clouds. No two clouds are the same, and there

When you look up into the sky, you will often see clouds. No two clouds are the same, and there are many different types of clouds. Different kinds of clouds lead to different types of weather. Cirrus clouds are the highest group of clouds that

271 views • 5 slides
Dry Drains: Myth, Reality or impediment to water conservation. Professor J.A. Swaffield Emeritus

Dry Drains: Myth, Reality or impediment to water conservation. Professor J.A. Swaffield Emeritus

Dry Drains presentation, CIBW62 Dusseldorf, 2009 Dry Drains: Myth, Reality or impediment to water conservation. Professor J.A. Swaffield Emeritus Profesor School of the Built Environment, Heriot Watt University, Edinburgh, Scotland. 1

378 views • 36 slides
Invigoration of deep convection in polluted environments: myth or reality? Wojciech W. Grabowski

Invigoration of deep convection in polluted environments: myth or reality? Wojciech W. Grabowski

Invigoration of deep convection in polluted environments: myth or reality? Wojciech W. Grabowski Mesoscale and Microscale Meteorology Laboratory NCAR, Boulder, Colorado, USA Results to be discussed are from two papers: Grabowski, W. W., and H.

722 views • 56 slides
Perpetual Access: Myth and Reality CRL webinar 20 January 2016 What is perpetuity?

Perpetual Access: Myth and Reality CRL webinar 20 January 2016 What is perpetuity?

Perpetual Access: Myth and Reality CRL webinar 20 January 2016 What is perpetuity? Definitions of Perpetual Lasting for eternity Lasting for an indefinitely long time In effect or having tenure for an unlimited duration Continuing

719 views • 50 slides
FUEL SUBSIDY; MYTH OR REALITY PRESENTED BY HENRY OLUMIJI BOYO CONVENER NIGERIANS UNITED FOR

FUEL SUBSIDY; MYTH OR REALITY PRESENTED BY HENRY OLUMIJI BOYO CONVENER NIGERIANS UNITED FOR

CIVIL SOCIETYS STATE OF THE NATION AND AGENDA SETTING ROUNDTABLE THEME FUEL SUBSIDY; MYTH OR REALITY PRESENTED BY HENRY OLUMIJI BOYO CONVENER NIGERIANS UNITED FOR DEMOCRACY , SOUTHWEST CHAPTER VENUE: SOCIAL JUSTICE CENTER, DATE: JUNE 17

421 views • 9 slides
Single Sign-On: Myth vs Reality Mark Wilcox mark@mjwilcox.com My Involvement Chief

Single Sign-On: Myth vs Reality Mark Wilcox mark@mjwilcox.com My Involvement Chief

Single Sign-On: Myth vs Reality Mark Wilcox mark@mjwilcox.com My Involvement Chief Integration Geek for WebCT Wrote a book on LDAP Became expert on authentication Participant in Internet 2 authentication working groups

789 views • 19 slides
Leadership in a Multigenerational Workplace Bridging the Gap to Drive Results Tresha Moreland,

Leadership in a Multigenerational Workplace Bridging the Gap to Drive Results Tresha Moreland,

Leadership in a Multigenerational Workplace Bridging the Gap to Drive Results Tresha Moreland, SPHR, SSBBP Test Your Knowledge Myth or Reality? The millennial generation is the largest generation of today Myth or Reality? Only older

505 views • 35 slides
INDEPENDENCE OF CENTRAL BANKS: MYTH OR REALITY? Experience of the Central Bank of Congo (BCC)

INDEPENDENCE OF CENTRAL BANKS: MYTH OR REALITY? Experience of the Central Bank of Congo (BCC)

INDEPENDENCE OF CENTRAL BANKS: MYTH OR REALITY? Experience of the Central Bank of Congo (BCC) Experience of the Central Bank of Congo (BCC) Deogratias MUTOMBO MWANA NYEMBO Governor Malabo, August 2015 CONTENTS I. OVERVIEW OF BCC

172 views • 16 slides
Durable Powdery mildew resistance in grapevines: myth or reality Summaira Riaz University of

Durable Powdery mildew resistance in grapevines: myth or reality Summaira Riaz University of

Durable Powdery mildew resistance in grapevines: myth or reality Summaira Riaz University of California, Davis Photo: NMSU Photo: Peter Magare Integrative approach Identification of resistance and marker development Durable Pathogen

426 views • 14 slides
Visual Debugger for Jupyter Notebooks: Myth or Reality? Elizaveta Shashkova EuroPython 2019

Visual Debugger for Jupyter Notebooks: Myth or Reality? Elizaveta Shashkova EuroPython 2019

Visual Debugger for Jupyter Notebooks: Myth or Reality? Elizaveta Shashkova EuroPython 2019 About Me Software Developer at JetBrains, PyCharm IDE Debugger and Data Science tools @lisa_shashkova 2 Visual Debugger 3

892 views • 68 slides
Timing issues in desktop audio playback infrastructure Alexander Patrakov April 11, 2015 About

Timing issues in desktop audio playback infrastructure Alexander Patrakov April 11, 2015 About

Timing issues in desktop audio playback infrastructure Alexander Patrakov April 11, 2015 About myself I am not working for any audio or open-source company I have submitted some PulseAudio patches I wrote dcaenc I added a

645 views • 40 slides
OSiRIS: A Distributed Storage and Networking Project Update Open Storage Research Infrastructure

OSiRIS: A Distributed Storage and Networking Project Update Open Storage Research Infrastructure

OSiRIS: A Distributed Storage and Networking Project Update Open Storage Research Infrastructure Shawn McKee for the OSiRIS Collaboration University of Michigan, Indiana University, Michigan State University, Wayne State University UM Storage

325 views • 28 slides
IEEE 802.15.4 and Zigbee CS 687 University of Kentucky Fall 2015 Acknowledgment: Some slides

IEEE 802.15.4 and Zigbee CS 687 University of Kentucky Fall 2015 Acknowledgment: Some slides

IEEE 802.15.4 and Zigbee CS 687 University of Kentucky Fall 2015 Acknowledgment: Some slides are adapted from presentations by Bob Heile from ZigBee Alliance, Joe Dvorak from Motorola, Geir E. Oien from NTNU, and Marco Naevve from Eaton

480 views • 34 slides
Ethernet transport protocols for FPGA Wojciech M. Zabootny Institute of Electronic Systems,

Ethernet transport protocols for FPGA Wojciech M. Zabootny Institute of Electronic Systems,

Ethernet transport protocols for FPGA Wojciech M. Zabootny Institute of Electronic Systems, Warsaw University of Technology Previous version available at: https://indico.gsi.de/conferenceDisplay.py?confId=3080 Joint CBM/Panda DAQ

359 views • 32 slides
Objectives An ASIC application MSDAP Analyze the application requirement System

Objectives An ASIC application MSDAP Analyze the application requirement System

Objectives An ASIC application MSDAP Analyze the application requirement System level setting of an application Define operation mode Define signals and pins Top level model Write a specification When addressing

681 views • 53 slides
Welcome! Todays Agenda: Self-modifying code Multi-threading (1)

Welcome! Todays Agenda: Self-modifying code Multi-threading (1)

/INFOMOV/ Optimization &amp; Vectorization J. Bikker - Sep-Nov 2019 - Lecture 13: Snippets Welcome! Todays Agenda: Self-modifying code Multi-threading (1) Multi-threading (2) Experiments INFOMOV Lecture 13

446 views • 43 slides
Linux perf_events status update Stephane Eranian Google Petascale Tools Workshop 2016 Agenda

Linux perf_events status update Stephane Eranian Google Petascale Tools Workshop 2016 Agenda

Linux perf_events status update Stephane Eranian Google Petascale Tools Workshop 2016 Agenda New hardware support Advanced Skylake PMU features Jitted code support A little Quizz New hardware support Intel Broadwell server

863 views • 20 slides
Evaluating and improving kernel stack performance for datagram sockets from the perspective of

Evaluating and improving kernel stack performance for datagram sockets from the perspective of

Evaluating and improving kernel stack performance for datagram sockets from the perspective of RDBMS applications Sowmini Varadhan(sowmini.varadhan@oracle.com) Tushar Dave(tushar.n.dave@oracle.com) Agenda What types of problems are we

591 views • 25 slides

More recommend