it press tour #19 slides openio june 2016
TRANSCRIPT
Application-awarestorage
@LaurentDenel
Co founder & CEOat OpenIO
@GuilDelaporte Co founder & Product Managerat OpenIO
Agenda
2.
OpenIOWebUI
Storage Tiering
3.1.
What is OpenIO
4.
Kinetic Drives
5.
Hybrid Cloud with Backblaze B2
1. What isOpenIO
2006
Idea & 1st concept
2007
Designdev starts
2009
1st massiveproduction above 1PB
2012
Opensourced
2014
10+ PBmanaged in asingle cluster
About
OpenIOfork
2015
20 60employees millions end-users
San Francisco | Lille (FR) | Tokyo | Montreal
Use cases
> Email Platforms > Video Streaming > Object and file storage
• Storage-as-a-Service • Compute + Storage Platform • On-prem Data
> High frequency modifications
most
88 % and growing fast
iops
capacity
> Low frequency modifications & immutable data
Storage market evolution (80/20 rule)
IOPS Realm Hyper-Scale Capacity Realm
x86 servers Software-Defined Storage
Hyper ScalableStorage and Compute Platform
The OpenIO Solution
+ =
OpenIO transforms a rack of x86 servers into a large storage and compute pool
OpenIO Democratizes Large Data Platforms
Internet giants have initiated the wave and proved the technology
OpenIO offers the same model “on-premise “
> Simple > Elastic > Flexible > Cost effective > On-demand
OpenIO Approach
Object Storage
Scale and store 1000+ PBs of data, Billions of objects
Open Source softwareCommodity hardware
Reduce cost and TCO
Application-aware Grid for Apps
Run applications on the storage infrastructure
1. 2. 3.
HOW
…
……
Multi tenancy
Flat structure
Track containers, not objects
Directory with indirections
Containers Objects
grid://namespace/account/container/object
>
Namespace
Account
Container
Object
New nodes are automatically discovered and immediately available
Scaling
No consistent hashing algorithm > no recalculation of the key space
Grid of nodes with no consistent hashingNever rebalance>
1. Collects metrics from the services of each node
2. Computes a score for each service
3. Distributes scores to every node and client
4. On the fly best match making for each request
The score is computed with a configurable formula usually based on: capacity, io performance, CPU
ConscienceRealtime load balancing for optimal data placement
>
Grid for AppsData usage at the heart of the datacenter >
A data processing framework integrated inside OpenIO’s Grid
Scale-out application back-ends can be built on the storage platform itself
Avoid wasted resources and simplify load balancing for storage and processing
APIs and Specific App ConnectorsData at the heart of the datacenter >
Gateway layer
Optimized native APIs
• C • Python • Java • Go
Command line interface
Specific App Connectors
• Enterprise Storage • Email • Video streaming
REST APIs
• Amazon S3 • Openstack Swift
What makes us unique?
3.2.1.Ease of use
Easy to test, deploy in production, manage and use
0 TB > 1000+ PB
Start small and grow with your needs (from one engine for small config to very large ones)
Grid for Apps
Unifying Compute and Storage on a single platform
1. Grid for Apps
Ingest Search
Store
Full Text Index
Video workflow for user generated contents
Ingest Stream
Store
Transcode
Store
Spam sample long-term archiving and search
Scale-out real life use cases
2. From 0TB to 1000+PB
Start small! No compromise on 3 object storage characteristics
Fast
> Simple and standard APIs
> Data protection with replication or erasure coding on multiple nodes
> Scale-out capabilities to hyper-scale
> A 3 nodes cluster can be deployed in 5 minutes and be production ready
3. Ease of use
Full Operational Control
CLI available
Ubiquitous Management via Web GUI
2. OpenIO Web UI
What you can do in 5’
Drink an espresso
Stretch your body
Deploy an OpenIO cluster
3.Storage Tiering
OpenIO tiering is obvious
1.Built-in feature
Across storage devices managed by OIO or connected to an external pool
2.Multicriteria settings
Name, date, size, patterns, type, container name, etc.
3.Transparent
At upload, or with background jobs
OIO chunk Layer
OIO directory
Client side applications
OpenIO standard architecture
R R R R R R R R
Storage Array
x86 Commodity Servers
EthernetTCP/IP drives
1995 - 2005 2006 - 2016 2017 - …
Large capacity storage evolution - drive presentation cost cutting
4. Kinetic Drives
OpenIO & Kinetic
Thanks Seagate for the Ember system!
KineticOpen SourceProject Member
Participation to the Plugfest in RaleighApril ‘16
Working proof of concept in our lab
Ready to get certified during the next Plugfest
github.com/open-io/oio-kinetic
OpenIO standard architecture
OIO chunk Layer
OIO directory
Client side applications
R R R R R R R R
OpenIO Kinetic architecture NO CHANGE!
OIO Kinetic Layer
K K K K K K K K
OIO directory
Client side applications
5. Hybrid Cloud
OpenIO with Backblaze
June 2016 Gleb Budman, CEO
WHEN WHO
$5/Month = 30GB
“Backblaze Undercuts All Cloud Storage Competitors”
Key Metrics
30.000 developers & IT people signed up
200+ PB stored
Profitable Hundreds of PB in the B2 pipeline
IT struggling with archiving data explosion
1. Explosion of Data Volume
2. Complexity
3. Infrastructure cost
Current solution: Silos
Hot data stored on disk
Cold data stored on LTO
Archiving softwareMove cold data to cold storage
But the current solution has lots of problems
Backblaze B2 & OpenIO take up the challenge
Data is siloed LTO has very slow performance
High cost of having multiple solutions
1. 2. 3.
A hybrid cloud storage solution to rule them all
OpenIO single namespace & tiering
Backblaze B2public cloud No more "cold" storage
Backblaze B2 offered through OpenIO solution
Data is siloed LTO has very slow performance
High cost of having multiple solutions
1. 2. 3.
1. OpenIO single namespace & tiering
OIO chunk Layer
OIO directory
Client side applications
R R R R R
Blackblaze B2 objects
HybridCloud
GW
2. Backblaze B2 public cloud / No more "cold" storage
2.
High performance
Unlimited storage
3.1.
Data is instantly available
4.
Pay only for what you use
3. Backblaze B2 offered through OpenIO solution
OpenIO (on premise)
Backblaze B2 (cloud)
hybrid
Push your datacenter’s walls!
2. Fast Performance
3. 99.999999% durability
Backblaze B2 Fully integrated in OpenIO
1. Cost Effective
Only $.005/GB/month
No Capex Pay for what you use
All data are stored on in-production. Always online disks. Parallelization achieves fast throughput.
All data checked ever 30 days for degradation and rebuilt when necessary.
3 years TCO for 1PB
Legacy OpenIO + Backblaze B2
Storage array
3 years hosting
Tape gear
Backup software
Human admin
OIO subscription
x86 servers
3 years hostingHuman adminB2
(>40% archived in Backblaze B2)
do you still want to live in the ’80s?
Ease of Use
3.
5 minutes deployment
Kinetic Drives
4.
Ready for the next storage hardware
2.
Tiering
Built-in, multi criteria and transparent
Grid for Apps
1.
scale-out backend for applications
Wrapping up
Backblaze B2
5.
Unique hybrid cloud storage approach
Application-AwareStorage