Beyond EBS Stroage Alternatives in the Cloud
Download as PPTX, PDF0 likes229 views
This document discusses the evolution of Amazon EC2 and EBS cloud storage offerings over time from 2006 to 2012. It notes limitations of early offerings like local instance storage and EBS in terms of performance, durability, and suitability only for basic applications. More recent additions like Provisioned IOPS EBS improved performance but gaps remain compared to traditional enterprise storage. The document argues for a new generation of block storage for the cloud that provides independent scaling of performance and capacity, guaranteed quality of service, higher durability and availability, efficiency through data reduction, automation, and true cloud scale. It suggests all-flash storage designed for cloud providers could help by restoring balance between performance and capacity while reducing costs of storage infrastructure.
Beyond EBS Stroage Alternatives in the Cloud
- 1. Advancing the way the world uses the cloud Beyond EBS – Storage Alternatives in the Cloud Dave Wright – Founder & CEO Slides @ http://www.slideshare.net/SolidFireInc
- 2. Margin High Med Low IOPS Applications Test / Development Backup / Archive Basic Web Cloud Opportunity Why is block storage so important? Phase 1 Phase 2 Performance Sensitive Apps Oracle / SAP / SQL Hadoop / NoSQL Complex Web apps Cloud Evolution $$$ $$ $ IaaS Cloud Today Key to bridging the gap: • Better block storage
- 3. Better block storage offerings critical to migrating high-margin apps from on-premises, managed, and dedicated environments to the cloud
- 4. A Brief History of Amazon EC2 & EBS
- 5. 2006 EC2 Launched • Storage options limited to S3 (object storage) and local instance storage • “Ephemeral” – no persistence • Limited performance – “SATA-like” IOPS and bandwidth – Shared resource (on smaller instances) • No HA – lose an instance, lose your data • Fixed capacity – tied to instance size • Only suitable for custom apps and basic web serving
- 6. 2008 EBS Launched • Network attached, persistent storage • Easily move between instances (in an Availability Zone) • More flexible capacity – 1GB 1TB • Poor performance – ~100 IOPS, Highly variable latency • Only for extra data volumes – boot volume still ephemeral • Failure rate of 0.1-1% (annual) (2 nines) • Snapshot backup/restore to S3
- 7. 2009 Boot from EBS • Added ability to store boot volumes on EBS • Finally allows EC2 instances to be provisioned and used like a “persistent” resource • Suitable for deploying some persistent, enterprise applications – If limited performance, durability, and capacity are acceptable
- 8. 2012 Provisioned IOPS • Added ability to set performance for an individual volume • Individual volume performance of 100 – 1000 IOPS (fixed at creation time) • (Likely) SSD based • Designed for 99.9% performance delivery (but no SLA) • Suitable for a wider range of performance-sensitive databases and applications
- 9. Evolution of EBS Ephemeral volumes Developer- oriented No enterprise apps Compute centric EBS Basic persistent storage Dev/test and limited production apps Low IO DBs/apps Boot from EBS Fully persistent virtual machines Basic enterprise apps Provisioned IOPS Controlled Performance Wider range of performance sensitive apps
- 10. Where is the rest of the cloud industry? Object Storage Atmos, Nirvanix, Scality, Swift Backup, archival, content repository Cheap and deep Instance Storage Local shared disk Typically persistent No HA, limited performance Network Storage EMC, 3par, Netapp, Dell Persistent, HA Variable performance, No SLAs, limited scale
- 11. The rest of the industry is 3-5 years behind Amazon EBS, working with inferior technology designed for the Enterprise
- 12. Traditional Enterprise Storage doesn’t work in the cloud Performance • Unable to manage performance independent of capacity • Can not guarantee storage performance Efficiency • Low and inefficient utilization rates • Lack of high performance in-line data reduction Management • Complex manual management that lacks automation • Limited or Complex DR configuration Scale • Limited scalability of both capacity and performance • Manage multiple islands of storage
- 13. Catching up (and passing) EBS • More control over performance • Higher durability & availability • Greater efficiency (profit ↑, prices ↓) • Complete storage automation • Cloud scalability
- 14. Control over performance • Manage volume performance and capacity independently • Guaranteed QoS / sustained performance to every volume • Dynamic adjustment of provisioned IOPS • Create and bill against multiple tiers of performance • Isolate noisy neighbors • Individual volumes delivering 10k IOPS or more
- 15. Higher durability & availability • 4 - 5 9’s required for enterprise apps • Remote site replication for DR • Snapshot & backup support • Basic RAID insufficient – Long rebuild times – Significant performance penalty – Single points of failure G A A B B C C D D C B E E F FH C H D G G H H B
- 16. High efficiency • High capacity utilization - 85% • Data Reduction in-line – De-duplication – Compression – Thin Provisioning • Executed without performance impact • Drastic reduction in power, cooling, and floor space
- 17. Complete management automation • Simple REST-based APIs for storage • Multi-tenant provisioning, reporting, and performance isolation • Full feature set – backup, DR, snapshots, etc. • Comprehensive reporting support & billing integration • Cloud platform integration
- 18. Cloud Scale • Scale-out rather than scale- up architectures • Easy to add both capacity and performance • Support for petabytes of capacity, millions of IOPS, and tens of thousands of volumes • Avoid islands of storage and manual data management
- 19. What role can Flash (SSD) play in improving block storage in the cloud?
- 20. Flash Benefits for Block Storage • Performance – Restore balance between IOPS and Capacity – Better performance with varied workloads • Efficiency – Remove performance as a limiter on utilization – Reduce storage space, power, and cooling • Management – Better reliability – Easier to isolate workloads on shared media
- 21. What’s the downside? • Cost – Partially a perception issue – Can’t be ignored – cloud all about value • Existing Storage Architectures – Ability to get full performance from flash – Write amplification & endurance issues – Inconsistent performance in tiered storage
- 22. All-SSD storage solution designed specifically for cloud service providers • Clustered Scale-Out Architecture • Industry-Standard: Hardware and 10 GigE iSCSI • Massive scale: 100 nodes, 2PB, 5M IOPS • Cost Effective: Usable $/GB similar to traditional SANs
- 23. 1620 Pearl Street, Boulder, Colorado 80302 Phone: 720.523.3278 Email: [email protected] www.solidfire.com Questions?
Editor's Notes
- #7: Test & throw away volumes