opennebula techday boston 2015 - future of information storage with iss supercore and ceph

Future of Information Storage with ISS SuperCore and Ceph

Intelligent Systems Services Inc.

Alex Gorbachev, President

Neal Purchase, Solutions Architect

Cindy Markee, Director of Technology Sales

Email: supercore@iss-integration.com

Data Storage Challenges

• Always. Need. More. GB TB PB EB

• RAID rebuild times now takes days per disk

• Limits on SAN array expansion (once you hit the physical limits you have to buy more arrays)

• Cost. Cost. Cost.

• Have to choose: Speed? Cost? Size?

• Upgrades mean downtime

Intro to Ceph

• Created as a Ph. D. thesis by Sage Weil in 2007

• Method for storing data that does not depend on parity calculations, controllers or lookup tools

• Pseudo Random Data Distribution

• CRUSH – Controlled Replication Under Scalable Hashing

• Getting too weird?

Hotel with a Billion Rooms

Pseudo Random Hashing

Abel

Baker

Charlie

• Data location is based on data itself

• Secondary (Tertiary etc.) location computed on the fly

Ceph RADOS

• Reliable Autonomic Distributed Object Store

• Block Device store without limitations on size, scalability, performance

Why is this not everywhere?

• Focus on performance, features, scalability

• No SAN like interface

• Native drivers are not yet mainstream

• Cloud oriented rather than enterprise

• But RADOS is a mature, well performing technology, so we developed…

Client Network and Fabric

Virtualization Servers

SuperCore Delivery Cluster

SuperCore Active Storage Nodes

SuperCore Control-Monitor

iSCSI

Fibre Channel

NFS

CIFS/SMB

Standalone Servers

SuperCore Backup

Basic Node Architecture

Control and Management Interface

Why we think this is AWESOME…

Resilience and Survival

• Replaces the outdated RAID technology with distributed storage logic that (unlike RAID) delivers limitless expansion and instant rebuild capability.

• Core and delivery EACH have separate redundancy layers

• Self healing algorithm located and eliminates faults on the fly without limitations (RAID systems can take 1-2 hits before data gets irreversibly corrupted, while SuperCore can take many hits in different areas and remain functional).

Replication and Optimization

• Ability to set placement rules for storage (where does data go and how is it spread out to protect the most against failures - rack, row, data center, city, continent...)

• Highly available flash caching that can be individually configured for every volume and centralized tiered caching.

• Performance increases with growth vs. traditional systems

Economy

• Different client systems can share the same storage hardware while maintaining quality of performance and separation of resources

• Ability to control storage overhead to meet capacity and performance demands while maintaining the required degree of protection

• On-demand provisioning (Thin Provisioning) economizes storage use overall up to 40%

Questions?