Raspberry Pi 5: Challenges and Solutions in Bringing up an OpenGL/Vulkan Driv...
Object Storage
1. OBJECT STORAGE - A FRESH
APPROACH TO LONG-TERM
FILE STORAGE
Oskars Krastiņš
«Telia Latvija»
2.
3.
4.
5. STRUCTURED VS. UNSTRUCTURED DATA
2017-06-07
Structured Data
• High Degree of organization, such as a
relational database
Unstructured Data
• Information that is difficult to organize using
traditional mechanisms
Parameter Value
Patient James Brown
Date of Birth 02.13.1987
Complain 1 chest pain
The patient came in complaining of chest
pain, shortness of breath, and lingering
headaches…smokes 2 packs a day…
family history of heart disease…has been
experiencing similar symptoms for the past
12 hours
11. BLOCK (SAN) FILE (NAS) OBJECT
Transaction Units Blocks Files Objects, that is files with
custom metadata and
unique ID
Type The oldest, most basic
form of storage
Builds on top of block
storage
Abstracts file and block
Storing Stores data as blocks,
typically 512 bytes
Stores data as files,
typically in 4KB blocks
Stores data as objects,
typically in 1MB blocks
Protocols SCSI, Fiber Channel,
SATA
CIFS and NFS REST and SOAP over
HTTP
Supported type of update Supports in-place
updates
Supports in-place
updates
No in-place update
support; updates create
new object versions
Knowledge Has no knowledge of the
information it is storing –
context is all in
application layer
Has a hierarchical map of
files to blocks (paths),
and system metadata,
but no other knowledge
Has a flat namespace of
objects, managed by a
relational or key/value
database – may have
rich knowledge of objects
12. BLOCK (SAN) FILE (NAS) OBJECT
Best suited for Transactional data and
frequently changing data
(data bases, OS)
Shared file data
(everyday workload)
Relatively static file data
and as a cloud storage
(images, PDF’s, Video,
archives)
Purpose Best for IOPs intensive
workloads, because each
application IO is
consistent to the storage
block size
Middle of the road,
serves many different
workloads
Best for workloads with
intensive bandwidth and
large capacities
Strength High performance Simplified access and
management of shared
files
Scalability and distributed
access. No geographical
restrictions
Limitations Difficult and costly to
extend beyond data
center
Difficult and costly to
extend beyond data
center
Not recommended for
frequently changing data
transactions; doesn’t
provide a sharing
protocol with a locking
mechanisms
15. 2017-06-07
• Distributed database management system
• Designed for big data
• Schema-Free
• Scalable
• Fault tolerant
• No single point of failure - Peer to Peer architecture
• Has an SQL like query language
• NoSQL
NoSQL DATABASE
19. 4 U
90 HDD x 8 TB
720 TB x 10 servers
Total ~7PB RAW
20.
21. OBJECT STORAGE
• data lives directly with the object
• can be retrieved with a single API call without the
overhead associated with a relational database
• there is no scalability issue
TRADITIONAL
• relational database
• application to relate this data to a specific object
• expensive and challenging to scale
22. SIMPLICITY
2017-06-07
• PUT
• GET
• Unique object ID
• Immutable (write-once, read-many)
• Not available «modify in-place»
• Lock free
• Metadata
• Extremely scalable I/O performance
EB – Exabyte
According to IDC, the total amount of data storage world-wide will reach 133 exabytes by the year 2017, of which 80 percent will be required for unstructured data
http://emergingtechblog.emc.com/converged-infrastructure-big-data-storage-analytics/
content-addressable storage (CAS**)
Daži KB 100 līdz daži MB
30 milj grāmatu/dokumentu
not support the POSIX IO calls (open, close, read, write, seek)
database layer (called “gateway” or “proxy” services
maintains the map of an object ID to user-friendly metadata like an object name, access permissions
Immutable - nemainīgs