provisioning hardware for cluster applications
play

Provisioning Hardware for Cluster Applications 2 Friday, February - PowerPoint PPT Presentation

Sep 12, 2023 •292 likes •1.26k views

scc : Cluster Storage Provisioning Informed by Application Characteristics and SLAs Harsha V. Madhyastha*, John C. McCullough , George Porter, Rishi Kapoor, Stefan Savage, Alex C. Snoeren, and Amin Vahdat UC Riverside* and UC San Diego Bourns

  1. scc : Cluster Storage Provisioning Informed by Application Characteristics and SLAs Harsha V. Madhyastha*, John C. McCullough , George Porter, Rishi Kapoor, Stefan Savage, Alex C. Snoeren, and Amin Vahdat UC Riverside* and UC San Diego Bourns College of Engineering 1 Friday, February 17, 12

  2. Provisioning Hardware for Cluster Applications 2 Friday, February 17, 12

  3. Provisioning Hardware for Cluster Applications “Need 100 Queries/s” 2 Friday, February 17, 12

  4. Provisioning Hardware for Cluster Applications “Need 100 Queries/s” 2 Friday, February 17, 12

  5. Provisioning Hardware for Cluster Applications “Need 100 “Need 1000 Queries/s” Views/s” 2 Friday, February 17, 12

  6. Provisioning Hardware for Cluster Applications “Need 100 “Need 1000 Queries/s” Views/s” 2 Friday, February 17, 12

  7. Goals for Provisioning at Low-cost • High performance • Redundancy • Supporting multi-tenancy • High availability • ... Our focus: reach performance goal and minimize cost for a single application, emphasizing storage 3 Friday, February 17, 12

  8. Challenge: Large Configuration Space • Diverse server enclosures/architectures • Diverse storage options Size MB/s (r/w) IOPS Cost 7.2k-rpm 500GB 90/90 125/125 $213 15k-rpm 146GB 150/150 285/285 $296 SSD 32GB 250/80 2500/1000 $496 DRAM 1GB 13k/13k 1.6B/1.6B $36 4 Friday, February 17, 12

  9. Challenge: Large Configuration Space • Diverse server enclosures/architectures • Diverse storage options Size MB/s (r/w) IOPS Cost 7.2k-rpm 500GB 90/90 125/125 $213 15k-rpm 146GB 150/150 285/285 $296 SSD 32GB 250/80 2500/1000 $496 DRAM 1GB 13k/13k 1.6B/1.6B $36 • Current state-of-the-art: 4 Friday, February 17, 12

  10. Challenge: Large Configuration Space • Diverse server enclosures/architectures • Diverse storage options Size MB/s (r/w) IOPS Cost 7.2k-rpm 500GB 90/90 125/125 $213 15k-rpm 146GB 150/150 285/285 $296 SSD 32GB 250/80 2500/1000 $496 DRAM 1GB 13k/13k 1.6B/1.6B $36 • Current state-of-the-art: • Apply rules-of-thumb from experience • Trial-and-error with various configurations • Configuration duplicated to scale-out 4 Friday, February 17, 12

  11. Scale-out Shortcomings for Photo-sharing Application 5 Friday, February 17, 12

  12. Scale-out Shortcomings for Photo-sharing Application 4x 5 Friday, February 17, 12

  13. Scale-out Shortcomings for Photo-sharing Application 4x Insight : Match storage to workload, leverage heterogeneity 5 Friday, February 17, 12

  14. Goal: Understand Configuration Space • What is a low-cost configuration now? • What will low-cost configurations look like in the future? Model Measure Predict Application “in-the-small” “in-the-large” Performance 6 Friday, February 17, 12

  15. scc: Storage Configuration Compiler scc 7 Friday, February 17, 12

  16. scc: Storage Configuration Compiler Cluster Building-Blocks scc 7 Friday, February 17, 12

  17. scc: Storage Configuration Compiler Cluster Building-Blocks Application scc Model 7 Friday, February 17, 12

  18. scc: Storage Configuration Compiler Cluster Building-Blocks Application scc Model SLA Specification 7 Friday, February 17, 12

  19. scc: Storage Configuration Compiler Cluster Building-Blocks Application scc Model Cost SLA Specification SLA 7 Friday, February 17, 12

  20. scc: Storage Configuration Compiler Cluster Building-Blocks Application scc Model Cost SLA Specification SLA 7 Friday, February 17, 12

  21. Outline • Modeling Applications and Hardware • Computing low-cost configurations • Example • Validation • Applications of scc 8 Friday, February 17, 12

  22. Cluster Building Blocks • Many types of servers 9 Friday, February 17, 12

  23. Cluster Building Blocks • Many types of servers CPU • Cores 9 Friday, February 17, 12

  24. Cluster Building Blocks • Many types of servers CPU RAM • Cores • RAM 9 Friday, February 17, 12

  25. Cluster Building Blocks • Many types of servers CPU RAM • Cores • RAM • Storage HDD HDD 9 Friday, February 17, 12

  26. Cluster Building Blocks • Many types of servers CPU RAM • Cores • RAM • Storage HDD HDD • I/O & Network 9 Friday, February 17, 12

  27. Application Model • Breakdown application into: • Tasks (Computation) • Datasets (Storage) • Edges between Tasks and Datasets (I/O) • Edges among Tasks (dependencies) 10 Friday, February 17, 12

  28. Example Model: Photo-Sharing Datasets Tasks 11 Friday, February 17, 12

  29. Example Model: Photo-Sharing Datasets Tasks Photos Thumbnails (Tag, Photo) 11 Friday, February 17, 12

  30. Example Model: Photo-Sharing Datasets Tasks Photo Upload Photos Thumbnail Write Tags Thumbnails (Tag, Photo) 11 Friday, February 17, 12

  31. Example Model: Photo-Sharing Datasets Tasks Photo Upload Photos Thumbnail Write Tags Thumbnails (Tag, Photo) 11 Friday, February 17, 12

  32. Example Model: Photo-Sharing Datasets Tasks Photo Upload Photos Thumbnail Write Tags Thumbnails (Tag, Photo) 11 Friday, February 17, 12

  33. Example Model: Photo-Sharing Datasets Tasks Photo Upload Photos Thumbnail Write Tags Thumbnails View Photo (Tag, Photo) 11 Friday, February 17, 12

  34. Example Model: Photo-Sharing Datasets Tasks Photo Upload Photos Thumbnail Write Tags Thumbnails View Photo (Tag, Photo) 11 Friday, February 17, 12

  35. Example Model: Photo-Sharing Datasets Tasks Photo Upload Photos Thumbnail Write Tags Thumbnails View Photo (Tag, Photo) View Tag 11 Friday, February 17, 12

  36. Example Model: Photo-Sharing Datasets Tasks Photo Upload Photos Thumbnail Write Tags Thumbnails View Photo (Tag, Photo) View Tag 11 Friday, February 17, 12

  37. Example Model: Photo-Sharing Datasets Tasks 5ms Photo Upload Photos 100ms Thumbnail 10ms Write Tags Thumbnails 1ms View Photo (Tag, Photo) 10ms View Tag 11 Friday, February 17, 12

  38. Example Model: Photo-Sharing Datasets Tasks 5ms Photo Upload 1x200kB Photos 100ms Thumbnail 1x4kB 10ms Write Tags Thumbnails 10x1kB 1ms View Photo (Tag, Photo) 10ms View Tag 11 Friday, February 17, 12

  39. Example Model: Photo-Sharing Datasets Tasks 5ms Photo Upload 1x200kB Photos 100ms Thumbnail 1x4kB 1x200kB 10ms Write Tags Thumbnails 10x1kB 1ms 10x4kB View Photo (Tag, Photo) 10x1kB 10ms View Tag 10x1kB 11 Friday, February 17, 12

  40. Example Model: Photo-Sharing Datasets Tasks 5ms Photo Upload 1x200kB Photos 100ms Thumbnail 1x4kB 1TB 1x200kB 10ms Write Tags Thumbnails 10x1kB 1ms 20GB 10x4kB View Photo (Tag, Photo) 10x1kB 10ms View Tag 10x1kB 2GB 11 Friday, February 17, 12

  41. Photo-Sharing SLA Datasets Tasks Photo Upload Photos Thumbnail Write Tags Thumbnails View Photo (Tag, Photo) View Tag 12 Friday, February 17, 12

  42. Photo-Sharing SLA Datasets Tasks Photo Upload Photos Thumbnail Tasks/s Write Tags Thumbnails View Photo (Tag, Photo) View Tag 12 Friday, February 17, 12

  43. Outline • Modeling Applications and Hardware • Computing low-cost configurations • Example • Validation • Applications of scc 13 Friday, February 17, 12

  44. Navigating the Configuration Space Cost SLA 14 Friday, February 17, 12

  45. Navigating the Configuration Space Cost SLA 14 Friday, February 17, 12

  46. Navigating the Configuration Space Cost SLA 14 Friday, February 17, 12

  47. Navigating the Configuration Space Cost SLA 14 Friday, February 17, 12

  48. Navigating the Configuration Space Cost SLA 14 Friday, February 17, 12

  49. Guiding Principle to Meet SLA • Complex interaction across storage-type and dataset assignments • Need to consider costs of meeting SLA for each permutation • Our configuration space is: S • D datasets, S storage-types D configs 15 Friday, February 17, 12

  50. Meeting the SLA at Low-cost Photos Thumbs Tags 16 Friday, February 17, 12

  51. Meeting the SLA at Low-cost Photos 7.2krpm Thumbs 7.2krpm Tags 7.2krpm 16 Friday, February 17, 12

  52. Meeting the SLA at Low-cost Photos 7.2krpm Thumbs 7.2krpm Tags 7.2krpm # storage units per-dataset 16 Friday, February 17, 12

  53. Meeting the SLA at Low-cost Photos 7.2krpm Thumbs 7.2krpm Tags 7.2krpm # storage units per-dataset Assign CPUs for computation 16 Friday, February 17, 12

  54. Meeting the SLA at Low-cost Photos 7.2krpm Thumbs 7.2krpm Tags 7.2krpm # storage units per-dataset Assign CPUs for computation Pack into servers 16 Friday, February 17, 12

  55. Meeting the SLA at Low-cost Photos 7.2krpm Thumbs 7.2krpm Tags 7.2krpm # storage units per-dataset Assign CPUs for computation Pack into servers Compute cost 16 Friday, February 17, 12

  56. Meeting the SLA at Low-cost Photos 7.2krpm 7.2krpm 7.2krpm 7.2krpm 7.2krpm ... Thumbs 7.2krpm 7.2krpm 7.2krpm 7.2krpm 7.2krpm Tags 7.2krpm 15krpm SSD 7.2k+DRAM 15k+DRAM # storage units per-dataset Assign CPUs for computation Pack into servers Compute cost 16 Friday, February 17, 12

Recommend

Perceus A Cluster Provisioning and Management Toolkit Bill Strossman Bill Strossman What

Perceus A Cluster Provisioning and Management Toolkit Bill Strossman Bill Strossman What

Perceus A Cluster Provisioning and Management Toolkit Bill Strossman Bill Strossman What Perceus Is An open source cluster toolkit for provisioning and managing stateless or stateful clusters Highly scalable Highly flexible and

169 views • 14 slides
Introduction to PC-Cluster Hardware I Russian-German School on High-Performance Computer Systems,

Introduction to PC-Cluster Hardware I Russian-German School on High-Performance Computer Systems,

Introduction to PC-Cluster Hardware I Russian-German School on High-Performance Computer Systems, 27 th June - 6 th July, Novosibirsk 1. Day, 27 th of June, 2005 HLRS, University of Stuttgart Introduction to PC-Cluster Hardware I Slide 1 High

892 views • 45 slides
Using SAML and XACML for Complex Resource Provisioning in Grid based Applications Yuri

Using SAML and XACML for Complex Resource Provisioning in Grid based Applications Yuri

Using SAML and XACML for Complex Resource Provisioning in Grid based Applications Yuri Demchenko, Leon Gommans, Cees de Laat System and Network Engineering Group University of Amsterdam POLICY2007 Workshop 13-15 June 2007, Bologna Outline

532 views • 25 slides
Applications and Services Motivations, history 1 t E2E W 1st E2E Workshop in 2008 k h i 2008

Applications and Services Motivations, history 1 t E2E W 1st E2E Workshop in 2008 k h i 2008

3rd E2E Provisioning Workshop Prague, Czech Republic 29-30 November, 2010 Peter Szegedi, PDO szegedi@terena.org www terena org www.terena.org TERENA TERENA End-to-End (E2E) Provisioning Workshop series 3rd workshop on Applications and

540 views • 15 slides
Cluster-Based Scalable Scalability Linear increase in hardware to handle load Network

Cluster-Based Scalable Scalability Linear increase in hardware to handle load Network

Advantages of Clusters Cluster-Based Scalable Scalability Linear increase in hardware to handle load Network Services Adding resources easy for clusters Availability 24 x 7 service, despite transient hardware or software errors

88 views • 4 slides
Provisioning IoT with Web NFC Zoltan Kis (@zolkis), Intel Background JavaScript APIs for

Provisioning IoT with Web NFC Zoltan Kis (@zolkis), Intel Background JavaScript APIs for

Provisioning IoT with Web NFC Zoltan Kis (@zolkis), Intel Background JavaScript APIs for IoTivity, Soletta W3C Web NFC editor Web access to hardware Earlier: Network management (DSL) Mesh radio networks DSP (AI applied in

908 views • 39 slides
Cluster algebras and applications Bernhard Keller Universit Paris Diderot Paris 7 DMV

Cluster algebras and applications Bernhard Keller Universit Paris Diderot Paris 7 DMV

Preliminaries: The Dynkin diagrams Definitions: quiver mutation, cluster algebras Application in Lie theory, after B. Leclerc et al. Application to discrete dynamical systems: periodicity Cluster algebras and applications Bernhard Keller

797 views • 22 slides
Resource provisioning requirements are very diverse Existing resource provisioning solutions

Resource provisioning requirements are very diverse Existing resource provisioning solutions

Resource provisioning requirements are very diverse Existing resource provisioning solutions tend to be specialized for specific scenarios. There is no resource provisioning model that supports all of these usage scenarios simultaneously

397 views • 14 slides
The QoE Provisioning-Delivery- g y -- Hysteresis and its Importance for Service Provisioning

The QoE Provisioning-Delivery- g y -- Hysteresis and its Importance for Service Provisioning

NGI 2011 The QoE Provisioning-Delivery- g y -- Hysteresis and its Importance for Service Provisioning in the F t Future Internet I t t Tobias Hofeld, Markus Fiedler and Thomas Zinner Hysteresis Hysteresis Hysteresis refers to

251 views • 10 slides
Mesos Problem Different applications need different frameworks How can we share a cluster

Mesos Problem Different applications need different frameworks How can we share a cluster

Mesos Problem Different applications need different frameworks How can we share a cluster among multiple frameworks? Statically partitioning the cluster Centralized task scheduler Key ideas Fine-grained sharing Key ideas

205 views • 18 slides
TERENA TERENA End-to-End (E2E) Provisioning Workshop End to End (E2E) Provisioning Workshop

TERENA TERENA End-to-End (E2E) Provisioning Workshop End to End (E2E) Provisioning Workshop

GLIF Tech Winter meeting Hong-Kong, China 24 February, 2011 Peter Szegedi, PDO szegedi@terena.org www terena org www.terena.org TERENA TERENA End-to-End (E2E) Provisioning Workshop End to End (E2E) Provisioning Workshop series About

587 views • 25 slides
PerfMiner: Cluster-Wide Collection, Storage and Presentation of Application Level Hardware

PerfMiner: Cluster-Wide Collection, Storage and Presentation of Application Level Hardware

PerfMiner: Cluster-Wide Collection, Storage and Presentation of Application Level Hardware Performance Data Philip J. Mucci, Daniel Ahlin Lars Malinowski, Per Ekman, Johan Danielsson Center for Parallel Computers (PDC), Royal Institute of

880 views • 50 slides
Energy-Efficient ARM64 Cluster with Cryptanalytic Applications 80 Cores That Do Not Cost You an

Energy-Efficient ARM64 Cluster with Cryptanalytic Applications 80 Cores That Do Not Cost You an

Energy-Efficient ARM64 Cluster with Cryptanalytic Applications 80 Cores That Do Not Cost You an ARM and a Leg Latincrypt 2017, 21st September 2017 1/19 Thom Wiggers Outline Introduction Building a cheap cluster The Cortex-A53 Breaking ECC

741 views • 70 slides
Chip Watson Scientific Computing Group Quick Outline Hardware Overview & Recent

Chip Watson Scientific Computing Group Quick Outline Hardware Overview & Recent

Chip Watson Scientific Computing Group Quick Outline Hardware Overview & Recent Changes Operations Report 2012 Conventional Infiniband x86 Cluster 2012 Accelerated Cluster Plans Hardware Overview IB

556 views • 29 slides
The NIDS Cluster: Scalable, Stateful Network Intrusion Detection on Commodity Hardware Matthias

The NIDS Cluster: Scalable, Stateful Network Intrusion Detection on Commodity Hardware Matthias

The NIDS Cluster: Scalable, Stateful Network Intrusion Detection on Commodity Hardware Matthias Vallentin 1 , Robin Sommer 2,3, Jason Lee 2 , Craig Leres 2 Vern Paxson 3,2 , and Brian Tierney 2 1 TU Mnchen 2 Lawrence Berkeley National

409 views • 25 slides
Black Hat Europe 2009 Hijacking Mobile Data Connections 1 Mobile Security Lab Provisioning

Black Hat Europe 2009 Hijacking Mobile Data Connections 1 Mobile Security Lab Provisioning

Black Hat Europe 2009 Hijacking Mobile Data Connections 1 Mobile Security Lab Provisioning & WAP primer Forging Messages Demo: Remote provisioning Provisioning: Process and Issues Attack scenario and exploiting Final

713 views • 59 slides
High Performance Computing Cluster 1 1 HPCC Hardware 20 Compute nodes, 560 cores with room to

High Performance Computing Cluster 1 1 HPCC Hardware 20 Compute nodes, 560 cores with room to

High Performance Computing Cluster 1 1 HPCC Hardware 20 Compute nodes, 560 cores with room to expand All Nodes: Xeon Broadwell CPUs (2.6 GHz) 28 cores per node Infiniband EDR (Infinite Bandwidth) (Enhanced Data

219 views • 5 slides
Story Applications Hardware Software Key features Roadmap Summary Story Story Applications

Story Applications Hardware Software Key features Roadmap Summary Story Story Applications

Story Applications Hardware Software Key features Roadmap Summary Story Story Applications Hardware Software Key features Roadmap Summary Todays enterprise-grade firewalls have highly advanced function

489 views • 29 slides
Story Applications Hardware Software Key features Roadmap Summary Story Story Applications

Story Applications Hardware Software Key features Roadmap Summary Story Story Applications

Story Applications Hardware Software Key features Roadmap Summary Story Story Applications Hardware Software Key features Roadmap Summary A low-cost modular Ethernet test solution designed from the bottom

446 views • 31 slides
Story Applications Hardware Software Key features Roadmap Summary Story Story Applications

Story Applications Hardware Software Key features Roadmap Summary Story Story Applications

Story Applications Hardware Software Key features Roadmap Summary Story Story Applications Hardware Software Key features Roadmap Summary Vulcan generates stateful Ethernet traffic to analyze how firewalls,

578 views • 37 slides
Cluster at the MPI for biophysical Chemistry 4 Racks, Rittal air/water cooling, 30 kW cooling

Cluster at the MPI for biophysical Chemistry 4 Racks, Rittal air/water cooling, 30 kW cooling

Cluster at the MPI for biophysical Chemistry 4 Racks, Rittal air/water cooling, 30 kW cooling capacity each Cluster Hardware 80 Nodes : Supermicro Sys-1027GR Motherboard 128 GB RAM 250 GB Harddisk CPU 2x Intel Xeon E5-2620

453 views • 6 slides
Cluster Architectures Overview Cluster Computing The Problem The Solution The Anatomy

Cluster Architectures Overview Cluster Computing The Problem The Solution The Anatomy

Cluster Computing Cluster Architectures Overview Cluster Computing The Problem The Solution The Anatomy of a Cluster The New Problem A big cluster example The Problem Applications Cluster Computing Many fields have come

857 views • 47 slides
Story Applications Hardware Software Key features Roadmap Summary Story Story Applications

Story Applications Hardware Software Key features Roadmap Summary Story Story Applications

Story Applications Hardware Software Key features Roadmap Summary Story Story Applications Hardware Software Key features Roadmap Summary Chimera is a network impairment emulator that makes it easy to

400 views • 22 slides
Introduction to PC- Cluster Hardware (II) Russian-German School on High Performance Computer

Introduction to PC- Cluster Hardware (II) Russian-German School on High Performance Computer

Introduction to PC- Cluster Hardware (II) Russian-German School on High Performance Computer Systems, June, 27 th until July, 6 th 2005, Novosibirsk 1. Day, 27 th of June, 2005 HLRS, University of Stuttgart High Performance Computing Center

865 views • 37 slides

More recommend