Their goal was to deliver an new level enterprise class level of ... Open Source. â¢Enterprise support and quality. â¢
Reinventing Cloud Storage Services Through Innovation and Collaboration Key properties for a cloud storage system
Background
• A major Global Telco provider was looking to enter the Storage as a Service market • Their goal was to deliver an new level enterprise class level of service to their customers • After a thorough review of a wide range of storage technologies and vendors, the customer selected Amplidata to develop a storage platform to meet their vision • Based upon this experience the following presentation is a brief overview of some of the areas to consider when developing your own public or private cloud implementation 2
AmpliStor® and Himalaya™ are trademarks of Amplidata. All other trademarks are the property of their respective owners.
Amplidata Confidential
Considerations
• Requirements – – – – – –
Exabyte scalable cloud That has built-in high durability: 15+ nines With guaranteed data consistency At low cost To offer cloud storage services to millions of users Across the globe
• Where would you start?
3
AmpliStor® and Himalaya™ are trademarks of Amplidata. All other trademarks are the property of their respective owners.
Amplidata Confidential
Choosing Your Platform Open Source
• Free community support • Expandable • COTS hardware
Commercial Storage System
• Expensive support & consultancy • Invest in in-house competency • In-house testing and tuning • No guaranteed solutions / SLA
• Enterprise support • Expensive license and support and quality • Long term lifecycle • Locked-in proprietary HW / release management
Software Enabled • Enterprise support and quality • Long term lifecycle / release management • Inexpensive COTS certified HW
4
AmpliStor® and Himalaya™ are trademarks of Amplidata. All other trademarks are the property of their respective owners.
Amplidata Confidential
High Durability at Low Overhead Erasure Encoding vs. 3 copy cloud • High Data Durability • 15 nines durability • •
99.9999999999999% Chance of losing an object per year is 1 / 1.000.000.000.000.000
•
Amazon is 11 nines
• Dynamically tunable in Policy • •
Tune to 2,3,4,5,6,…19 failure protection Update policy on-line
• Lower overhead = lower TCO • Only 43% storage overhead •
3 Copy Cloud Store
18/6 3 GEO AmpliStor
20/4 1 site AmpliStor
Source data
1.00 TB
1.00 TB
1.00 TB
Raw disk
3.00 TB
1.71 TB
1.43 TB
Overhead
200%
71%
43%
Technology
for 4 failure protection
• Compared to 200% for 3 copy
*This calculation is based on 3TB disks, 5% annual failure rate (AFR), 1PB AmpliStor® and Himalaya™ are trademarks of Amplidata. All other trademarks are the property of their respective owners.
Data Consistency Don’t settle for eventual consistency
by Wyatt Lloyd, Michael J. Freedman, Michael Kaminsky, David G. Andersen, April 21, 2014 http://queue.acm.org/detail.cfm?id=2610533
6
AmpliStor® and Himalaya™ are trademarks of Amplidata. All other trademarks are the property of their respective owners.
Amplidata Confidential
3D Elastic scalability
Scale global namespace
Scale throughput
Scale storage capacity
7
AmpliStor® and Himalaya™ are trademarks of Amplidata. All other trademarks are the property of their respective owners.
Amplidata Confidential
BitSpread® True 3D Elastic Scale-out ANY encoded block from ANY object in ANY bucket can be stored on ANY disk – Added disks become immediately available to store new data blocks – When disks are added, the existing data does not need to be moved – Buckets are NOT limited to subset of storage nodes – Spread NOT determined by physical installation of storage nodes – Hierarchy aware: spread wide to minimize impact of rack / DC failure Other technologies and limitations BitSpread Hierarchy aware spread 3D Elastic Scale-out
RAIN RAID pool across Storage Nodes Fixed hardware layout Expansion requires fixed # nodes Repair to same nodes in order
RING / CRUSH map Mapped to hardware layout Data movement at HW change
3 replica map 18/3 Pool
TRUE Exabyte Global pool
Copyright © Amplidata 2011-2014 8
AmpliStor® and Himalaya™ are trademarks of Amplidata. All other trademarks are the property of their respective owners.
6/2 Pool 1 geo Pool
6/1 Pool 3 geo Pool
Amplidata Confidential
12/4 EC map
6/2 EC map
GeoSpread® Multi-Site Disaster Prevention •
GeoSpread = hierarchy aware and configurable policy – WAN latency optimized consistency – No need to match exact hardware deployment to achieve Geo-protection
•
Protects against a full datacenter outage and more – Policy flexibility: DC failure + simultaneous disk/node/rack failure(s)
•
Local read performance – Automatically reads more data locally at any location, no configuration 1PB USER DATA Traditional Storage •
18/6 policy GeoSpread
•
3 replica + 25% RAID-6 overhead
•
1.71PB total RAW capacity
•
3.75 PB total RAW capacity
.57PB
+
.57PB
+
.57PB
1.25PB
+
1.25 PB
+
1.25PB
*Policy such as 18/7 can be set to protect against one datacenter + an additional node or rack in a second datacenter.
DC1
9
DC2
AmpliStor® and Himalaya™ are trademarks of Amplidata. All other trademarks are the property of their respective owners.
DC3
DC1
Amplidata Confidential
DC2
DC3
Small File Support
Requirement: •
Any bucket supports optimization for small files
•
•
10
Not acceptable: •
Configurable file size threshold that automatically optimizes for small files within bucket
The need to modify applications to redirect small files to different buckets
•
Same durability policy for small and large files: same failure protection level
Replication with low durability for small files vs. high durability for large files
•
Some systems don’t offer any protection at all for small files: they can be lost and might need to be regenerated
•
Dedicated hardware required for small files
•
No dedicated/specific hardware required for small files
•
High read performance, eliminates high number of disk IOPS for small files
AmpliStor® and Himalaya™ are trademarks of Amplidata. All other trademarks are the property of their respective owners.
Amplidata Confidential
The Customer solution: Introducing the Himalaya Reference Architecture for Service Providers A software defined object storage solution architecture Designed to meet true telco class service levels on a global scale Scalability
• 100+ Trillions of objects global namespace capacity • 100+ of Exabytes capacity
Durability
• Flexible policy definition supports multiple durability levels in a single system and online policy updates • True data consistency • Built-in robust data integrity, audit and repair functions
Performance
• 400+ MB/sec single stream performance • Linear performance scalability • Small file performance optimization
Security
• Network SSL/TSL and AES256 data encryption • LDAP authentication
Manageability
• Hot swap for HDD/SDD • HDD fail-in-place without degraded performance
Platform support
• Runs on Intel-based, Commercial-Off-The-Shelf hardware • Several suppliers certified
AmpliStor® and Himalaya™ are trademarks of Amplidata. All other trademarks are the property of their respective owners.
Amplidata Confidential
12
AmpliStor® and Himalaya™ are trademarks of Amplidata. All other trademarks are the property of their respective owners.
Amplidata Confidential
Service Provider vs. Enterprise Requirements Service provider •
Enterprise
Scale: Exabytes in multi-PB blocks
•
Great Out of the box experience
•
Self managing: little knowledge and infrequent operator intervention required
•
Broad API compatibility
•
Fully locked security
•
Storage operation API's
•
Start small, scale large in-place
•
User operation API’s self service portal, metering
•
Integration with account/authentication system
•
Bulk customer on-boarding
•
Certified Partner Ecosystem
•
Broad COTS HW support
•
Broad COTS HW support
Himalaya™ Reference Architecture available in 2 products: - Himalaya Service Provider/OEM - Himalaya Enterprise 13
AmpliStor® and Himalaya™ are trademarks of Amplidata. All other trademarks are the property of their respective owners.
Amplidata Confidential
Amplidata offering
Stay tuned for joint customer announcement tomorrow
AmpliStor® and Himalaya™ are trademarks of Amplidata. All other trademarks are the property of their respective owners.
Amplidata Confidential
Summary • When designing a public or private storage cloud, there are many things to consider. • The main goals should be to deliver: enterprise/telco class service durability and performance levels elastically 3D scale to meet capacity, performance and namespace requirements now and in the future a software defined solution that contains cost by supporting COTS hardware strong data consistency to make the storage easy to consume by users and applications 15
AmpliStor® and Himalaya™ are trademarks of Amplidata. All other trademarks are the property of their respective owners.
Amplidata Confidential
Himalaya Delivers 3D Elastic Scalability
Scale Throughput
Scale Global Namespace • •
Created new “Sharded” database architecture Scale by simply adding SSDs to system
•
•
Add Scale Servers Add Controllers
Scale Storage Capacity • •
Add Storage Nodes Storage Nodes will maximize different drive sizes
Plug & Play design allows system to scale linearly as hardware is added.
17
AmpliStor® and Himalaya™ are trademarks of Amplidata. All other trademarks are the property of their respective owners.
Amplidata Confidential
Customer Solution: Himalaya™ Reference Architecture
• A software defined object storage solution architecture • Massively scalable • Enterprise class performance • Flexible to meet variable SLA levels • Runs on Intel-based, Commercial-Off-The-Shelf hardware • Himalaya™ Reference Architecture available in 2 products – Himalaya Service Provider/OEM – Himalaya Enterprise
18
AmpliStor® and Himalaya™ are trademarks of Amplidata. All other trademarks are the property of their respective owners.
Amplidata Confidential