Scale Up is History! Is Scale Out the Future for Storage

Views:
 
     
 

Presentation Description

No description available.

Comments

Presentation Transcript

Contact us:

Scale Up is History! Is Scale Out the Future for Storage? A Technical and Market Analysis Chris M Evans Langton Blue Ltd Copyright © 2014 Langton Blue Ltd 1

Contact us:

About Our Speakers Chris M Evans 27 years experience in the IT industry across all platforms, including IBM mainframe, Windows & Open Systems. Co-founder and independent consultant at Langton Blue Ltd, a specialist consulting company in the UK. Blogger and part-time analyst at Architecting.IT Twitter: @chrismevans, @ architectingIT , @langtonblue Web: www.architecting.it , www.langtonblue.com Copyright © 2014 Langton Blue Ltd 2

Contact us:

What You Will Learn How Scale-Out Solutions are Defined The differences and benefits between scale out and traditional scale-up solutions How scale-out helps deliver hyper-converged compute and storage solutions How to decide whether scale-out is right for you Vendor Roundup – hardware and software offerings in today’s marketplace. Copyright © 2014 Langton Blue Ltd 3

Contact us:

Scale Up Storage Defined The dominant storage architecture for the past 25 years Components all built into a single “monolithic” hardware device Managed as a single hardware device Incorporates expensive highly redundant components Originally used proprietary components (ASICs, backplanes, internal bus) although industry has moved towards commoditisation Designed for maximum availability and up-time Performance and capacity increased by adding more components (processor, adaptor cards, directors, disks) to the single device Copyright © 2014 Langton Blue Ltd 4 EMC VMAX

Contact us:

Scale-Out Storage Defined Copyright © 2014 Langton Blue Ltd 5 Systems built from clusters of hardware (and software) nodes Support File, Block and Object protocols Built from commodity components (e.g. 1U standard servers & disk drives Protect data through RAID, Mirroring and Erasure Coding techniques Failure of a disk/flash device or entire node is designed into the architecture System scales capacity & performance by adding nodes to the configuration Scale Out concept used as the basis for hyper-converged solutions SolidFire SF Series

Contact us:

Scale-Up/Scale-Out – Pets v s Cattle Copyright © 2014 Langton Blue Ltd 6 Pets are Named; nursed back to health and cared for when ill Cattle are numbered, if they get ill, they are “expendable”

Contact us:

Scale-Up/Scale-Out - Comparision Scale Out “Loosely Coupled” architecture Multiple nodes connected through Network backplane (e.g. Ethernet) Commodity components Tolerates disk and node failures Storage/compute loss is 1/n of node count Growth through adding nodes/servers with compute and storage capacity Copyright © 2014 Langton Blue Ltd 7 Scale Up “Tightly Coupled” architecture Controllers Dedicated backplane Proprietary components Designed for 100% uptime of all components BBU, UPS, multiple redundancy planes Growth through adding faster controllers & disk shelves

Contact us:

Scale Out Benefits Lower Cost than Scale-Up Commodity components & SDS products Deploy as software-only with VMs Simpler maintenance – e.g. node-based upgrades, no data migration Scale “on demand” Grow capacity & performance by the node Simple deployment process – rack and join node cluster Lower Risk New nodes don’t have to be plugged directly into the running infrastructure (like disk shelves) 1/n capacity lost on node failure – more nodes = less impact Rolling upgrade – add new (faster/bigger) nodes, decommission the old ones – no fork-lift upgrades Greater Flexibility Geo-dispersal of data with erasure coding Software-only solutions, no hardware tie-in or cost Copyright © 2014 Langton Blue Ltd 8

Contact us:

Scale Out Issues Node Balancing Can have surplus or shortage of compute/storage resources with fixed node configurations (solved with multiple node types & dynamic balancing) Recovery Rebuild Process Can have performance impact if entire node’s data has to be recreated (solved with prioritised /intelligent rebuild) – scale out rebuild can be fast Network Dependency Infrastructure is limited in write I/O throughput to network bandwidth (need to protect new data across > 1 node) Storage Efficiency Many solutions use replicas (e.g. RAID-1/10), can be expensive on disk capacity, especially with all-flash solutions Minimum Node deployment Hardware solutions usually require minimum of 3 nodes, usually recommended to be 4 Copyright © 2014 Langton Blue Ltd 9

Contact us:

Scale Out and Hyper-Converged Hyper-converged hardware vendors have used scale-out storage concepts within their products Nutanix (NDFS), Scale Computing (SCRIBE), VMware EVO:RAIL (VSAN) Scale out storage and scale out compute are a “natural fit” for each other Scale out storage software can be used to deliver hyper-converged solutions without buying a packaged solution StarWind Virtual SAN, VMware VSAN, ScaleIO OpenStack takes advantage of scale-out properties Swift (object), Ceph (block), GlusterFS (File) Copyright © 2014 Langton Blue Ltd 10

Contact us:

Choosing The Right Product What are my requirements? Protocols (block, file object) Geo-dispersal Features – data protection, hypervisor support Hardware or Software only? Is Scale-Out suitable for all of my data? Mix scale up and scale out technologies where appropriate What’s my Growth/Capacity requirement? Object typically 100TB+ deployments File & block, 10’s of terabytes Existing process/technology synergies Can I use commodity servers already in my environment? No extra support/maintenance cost “Mix & Match” storage and compute Great for software-only solutions Can I build out OpenStack or scale-out Compute Farm? Copyright © 2014 Langton Blue Ltd 11

Contact us:

Vendor Roundup - Hardware Copyright © 2014 Langton Blue Ltd 12

Contact us:

Vendor Roundup - Software Copyright © 2014 Langton Blue Ltd 13

Contact us:

entirely software-based, VM-centric virtual machine storage, which basically “mirrors” internal hard disk and flash memory between hypervisor hosts StarWind Virtual SAN is… before after eliminates any need for physical shared storage like SAS JBODs, iSCSI or Fibre Channel or NAS and seamlessly integrates into the hypervisor

Contact us:

Zero Hardware footprint Complete fault tolerance with just 2 servers No SSD / 10GbE / SAS JBOD required VM-Centric storage design Maximized storage performance StarWind Virtual SAN – hardware-less storage

Contact us:

Basic 2-node configuration

Contact us:

Thinking storage outside the box

Contact us:

Q&A Copyright © 2014 Langton Blue Ltd 18

Contact us:

Contact us Copyright © 2014 Langton Blue Ltd 19 Download StarWind Virtual SAN 30-Day Trial here: www.starwind.com/starwind-virtual-san/ Headquarters: 301 Edgewater Place, Wakefield , MA 01880 USA/Canada: 1-617-449-7717 EMEA and Asia: +44-207-193-67-27 sales@starwindsoftware.com info@starwindsoftware.com

authorStream Live Help