Upgrade to Pro — share decks privately, control downloads, hide ads and more …

2018: "Composable Infrastructure: What It Means For Your Datacenter"

2018: "Composable Infrastructure: What It Means For Your Datacenter"

DriveScale presentation introducing Composable Infrastructure and the DriveScale product. Bay LISA April 19, 2018

Tom Lyon

April 19, 2018
Tweet

More Decks by Tom Lyon

Other Decks in Technology

Transcript

  1. Legal Notices This presentation and all information contained herein (collectively,

    the “Presentation”) constitutes the proprietary and confidential information of DriveScale, Inc. (“DriveScale”) and/or its licensors, and is protected by U.S. and international copyright law. All rights reserved. You agree not to reproduce, retransmit, disseminate, sell, distribute, publish, broadcast, circulate or commercially exploit the Presentation in any manner without the express written consent of DriveScale. THE PRESENTATION IS PROVIDED “AS IS”, AND DRIVESCALE EXPRESSLY DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY, ACCURACY OF ANY THE CONTENT OR MATERIALS PROVIDED HEREIN, FITNESS FOR A PARTICULAR PURPOSE, AND NONINFRINGEMENT. IN NO EVENT WILL DRIVESCALE OR ITS LICENSORS BE LIABLE TO YOU OR ANYONE ELSE FOR ANY CONSEQUENTIAL, INCIDENTAL, SPECIAL OR INDIRECT DAMAGES (INCLUDING BUT NOT LIMITED TO LOST PROFITS OR DAMAGES THAT MAY RESULT FROM INCONVENIENCE, DELAY OR LOSS OF THE USE OF THE PRESENTATION) EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGES. 2 © 2018 DriveScale Proprietary & Confidential
  2. Composable Infrastructure: History §  Facebook – 2013 – Disaggregation and

    Next-Generation System Design, J.Taylor §  Intel – 2014 - Rack Scale Design(Architecture) –  software & architecture for partners, fabric agnostic, cloud focus §  Cisco – 9/2014 – UCS-M Modular, Composable System –  Up to 16 micro-servers in one 2U chassis, storage and network resources re- configurable via “System Link” ASIC §  HPE – 12/2015 – Synergy –  new generation blade server – up to 12 servers in 10U chassis, storage composability via SAS fabric 4 ©2018 DriveScale Inc. All Rights Reserved.
  3. Composable Infrastructure: Startups §  DriveScale: “Software Composable Infrastructure for Your

    Data Center” §  H3Platform: NVMe and GPU disaggregation §  Liqid: “On-Demand Composable Infrastructure” §  TidalScale: “Software Defined Servers” §  NVMe over Fabrics: –  Apeiron, Attala, Excelero, E8 Storage, Pavilion Data 5 ©2018 DriveScale Inc. All Rights Reserved.
  4. Conflicting uses of “Composable” §  Compose.io (IBM) – managed database

    services §  Composable Software – research / software engineering topic §  Composable Analytics, Inc. – full stack analytics platform §  Composable Systems, Inc. – software outsourcing §  HTBASE – “Composable Operating System” – composing higher level services, not bare metal 6 ©2018 DriveScale Inc. All Rights Reserved.
  5. The Fabric is of the Essence §  Ethernet (DriveScale) – 

    Ports on every server –  Lots of extra bandwidth in the rack –  Highly interoperable, high redundancy available, not fragile §  SAS (HPE Synergy) –  Poor scaling, storage only –  Good interoperability, high redundancy available, not fragile §  External PCI Express (Liqid, H3Platform) –  Poor scaling, high performance, fragile 7 ©2018 DriveScale Inc. All Rights Reserved.
  6. Fundamentally: §  Composable Infrastructure allows “late-binding”, and re-binding of CPUs

    to other resources: –  Disk –  SSD –  GPU –  Accelerators –  Memory (future) §  “Software Defined Sheet Metal” §  Controlled via API 8 ©2018 DriveScale Inc. All Rights Reserved.
  7. Benefits of Composable Infrastructure §  Reduce over-provisioning §  Improve time-to-service

    §  Improve availability §  Simplify cluster management §  Prevent cluster silos §  Reduce SKU inventory 9 ©2018 DriveScale Inc. All Rights Reserved.
  8. DriveScale Leadership 11 Satya Nishtala, CTO •  Fellow, System Architect:

    Nuova (Cisco UCS) •  Lead architect: Sun - UltraSPARC workstations and work-group servers, Storage Tom Lyon, Chief Scientist •  Founder: Nuova (Cisco UCS) •  Founder: Ipsilon (Nokia) •  Emp #8: Sun Micro – SunOS, SPARC, SunScreen FOUNDERS Amr Awadallah Founder/CTO Cloudera Herb Cunitz President Hortonworks Whit Diffie Pioneer of Public Key Cryptograhy Scott McNealy Founder, CEO Sun Microsystems ADVISORS James Gosling Creator of Java Gene Banman, CEO •  CEO: ClearPower, Zero Motorcycles, NetContinuum •  VP/GM Desktops, Sun Micro •  President of Sun Japan Ingrid Burton CMO Hortonworks
  9. 12 IDC Forecast Data is Growing Exponentially Modern Workloads Traditional

    Workloads Big Data Drivers •  Big Data Analytics •  Social •  Mobile •  ML, AI •  IoT DriveScale Confidential & Proprietary 2018 Modern Workload Platforms
  10. vSphere Data Center Architecture Transition Required 13 ESX ESX ESX

    ESX Shared Storage – NAS or SAN VM App App App App VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM App App App App App App App App App App App App Traditional Workloads: Many Apps Managed on Each Server / Shared Storage Server Server Server Server Server Server Server Server Server Server Server Server Server Hadoop Modern Workloads: Large Apps Running on Clusters of Servers / Local Storage Cassandra X 100s or 1,000s Kubernetes DriveScale Confidential & Proprietary 2018 Server Server Server Server Server Server
  11. Slow response to changing workload demands Inability to balance resources

    between application silos Modern Workloads Present New IT Challenges 14 Poor utilization and/or excessive over- provisioning Server upgrades are expensive and cumbersome Provisioning times are too long and labor intensive Fixed ratio of compute-to-storage results in excessive spending and slow IT response times. No independent scaling of compute or storage
  12. DriveScale Composability Platform 15 Bare Metal DriveScale Composability Platform PCIe

    Ethernet iSCSI RoCE v2 NVME/TCP Rack Servers HDD SSD NVMe Blade Servers Modular Servers GenZ NoSQL
  13. DriveScale Software Composable Infrastructure 16 1) Disaggregate servers into… • 

    Compute resources (diskless servers) •  Storage resources (JBODs, storage arrays, NVMe SSDs) 2) Compose nodes and clusters by… •  Combining servers with storage via software •  Dynamically re-compose as needs change Two Simple Steps:
  14. Step 1: Disaggregate the Servers Typical Rack Server Configuration DriveScale

    Pooled Resources Diskless Servers DriveScale Adapter JBODs (Just a Bunch of Disks) Instead of buying servers with internal storage… Start buying diskless servers with external storage
  15. Step 2: Dynamically Compose Servers and Clusters DriveScale Pooled Resources

    Diskless Servers Easily compose software-defined physical nodes and clusters on-the-fly Combine just the compute and storage resources needed for each workload Easily add, remove or reallocate resources, even between clusters – no more silos! Performance equivalent to bare metal servers with direct attached storage No changes required to the application stack JBODs (Just a Bunch of Disks) DriveScale Adapter
  16. Scalable, Right-sized Clusters From Pools of Resources Cluster 1 Balanced

    Cluster 2 Storage intensive Cluster 3 Compute intensive Define clusters based on individual workload requirements Diskless Servers JBODs (Just a Bunch of Disks) DriveScale Adapter
  17. Scalable, Right-sized Clusters From Pools of Resources Define clusters based

    on individual workload requirements Clusters can stripe across racks and scale as needed Diskless Servers JBODs (Just a Bunch of Disks) DriveScale Adapter
  18. DriveScale Management System •  Collects inventory from Agents and Adapter

    •  Node and cluster configuration and composition •  Constraint management •  Node and cluster template creation •  Physical and logical node status Typical deployment: •  1-3 per customer •  Linux RPM; runs on VMs DriveScale Solution Software Components DriveScale Central •  Cloud-based portal for administration of all sites •  Download software updates, view log files, documentation, etc. DriveScale Server Agent •  Lightweight agent runs on Linux servers •  Provides server node inventory to DMS •  Includes server configuration and monitors status Typical deployment: •  Runs on each server node •  Can be automatically deployed DriveScale Agent
  19. Compute and Storage Vendor Agnostic 22 DriveScale works with your

    choice of compute and storage solutions Compute Options •  Rack Servers •  Blade Servers •  Modular Servers Storage Options •  SAS HDD •  SAS SSD •  Soon: NVMe SSD SAS JBOD w/ DriveScale Adapter Soon: Composable NVMe Systems
  20. DriveScale Adapter 23 §  High Performance SAS-Ethernet Bridge –  1RU

    appliance w/ dual power supplies –  4 controllers (2 x 10 GE, 2 x 12Gb 4 lane SAS) –  80 Gb throughput §  Leverages Ethernet for optimal performance and cost §  Multipath redundancy (4 paths from any server to any drive) §  Patented traffic balancing algorithms ToR Switches 10 GE Diskless Servers DriveScale Adapter SAS JBOD Card 4 Card 3 Card 2 Card 1 Node N Node 1 Controller 1 Controller 2 … SAS 10GE
  21. 24 Storage Type DAS (Direct Attached Storage) Software Composable Infrastructure

    Centralized Storage NAS or SAN Cost Performance Utilization Flexibility Scalability The best of both worlds! Software Composable Infrastructure is a better option The performance and cost of DAS with the utilization of NAS/SAN. DriveScale Confidential & Proprietary 2018
  22. DriveScale Does for Clusters what VMware Did for Servers 25

    ESX ESX ESX ESX External Storage – NAS or SAN vSphere VM App App App App VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM App App App App App App App App App App App App © 2018 DriveScale Proprietary & Confidential VMware Solved Server Utilization Inefficiencies and Flexibility Limitations DriveScale Software Composable Infrastructure DriveScale Solves Cluster Utilization Inefficiencies and Flexibility Limitations
  23. DriveScale Inc. 1230 Midas Way, Suite 210 Sunnyvale CA 94085

    www.drivescale.com Thank You. ©2018 DriveScale Inc. All Rights Reserved. 26