Network data storage

Содержание и основные понятия

Back to the table of contents
Alterna · Chapter 5 · page 5.3
Network data storage
Data can be stored locally, on a network device, or in a cloud.
Network vault
The data are not stored on your computer, but on a special server, and are available from any connected device.
From the NAS home to Google data centres, principle one: data are separated, duplicated and available online.
GEF · Basic
PLF · Infred
01
Network storage types
DAS, NAS, SAN - three approaches to data storage
GEF
DAS — Direct Attached Storage
Disk connected.
Directly
To the computer (internal SSD, external USB disc). Quickly, but the data are only available from this computer.
NAS — Network Attached Storage
Separate device connected to
Local area network
It works as a file server: connects to Ethernet, provides a common folder. Access from any device on the SMB (Windows), NFS (Linux), AFP (MacOS) protocol network.
SAN — Storage Area Network
Selected
high-speed network
It works at the block level (block-level) rather than the file level. It is used in major data centres. The protocols are Fibre Channel, iSCSI.
Type
Level of access
Protocols
Speed
Application
DAS
Blocks/ Files
SATA, NVMe, USB
Maximum
Personal PC
NAS
Files
SMB, NFS, FTP
Average (1-10 Gbit/s)
Small business, home
SAN
Blocks
Fibre Channel, iSCSI
High (16 - 128 Gbit/s)
Date centres, enterprises
🏠
Home NAS
analogy: shared closet in the apartment
Synology or QNAP with 2–4 discs.
🏢
Corporate SAN
Analog: Industrial warehouse
NetApp, Dell EMC. Hundreds of disks, double controllers, read speed, tens of GB/s.
02
RAID Masses
RAID 0, 1, 5, 10 - How to combine disks for speed or reliability
PIF
What's RAID?
RAID
(Redundant Array of Independent Disks) is a technology that consolidates several disks into one logical array to increase speed, reliability, or both.
Main RAD levels
Level
Principle
Min Disk
Capacity
Frustration
Speed
RAID 0
Creeding (stripe)
2
N × V
None
Maximum
RAID 1
Mirror Mirrors
2
V
1 disc
Reading x2, recording x1
RAID 5
Equivalence + evenness
3
(N−1) × V
1 disc
Good.
RAID 6
Equivalence + 2 evenities
4
(N−2) × V
2 discs
Good.
RAID 10
Mirror + stripe
4
(N/2) × V
1 to mirror
High
N
- the number of disks,
V
- the volume of one disk.
RAID 0 is a sequence
Data are broken down into blocks and recorded
in turn
The speed of reading/recording is doubled, but if
Any
The disk will fail, all the data is lost.
RAID 1 is a mirror.
Every byte is recorded
on both discs simultaneously
If one disc breaks, the data will remain on the second. Capacity = volume of one disk (second is a copy).
RAID 5 — Balance
The data rotates as in RAID 0, but is additionally calculated
evenness block
(XOR) It's distributed to all disks.
one
The disk is restored from the remaining + evenness.
RAID 10 is a mirror + stripe
The combination RAID 1 and RAID 0: Discs are broken into pairs and the data rotate between pairs. Quick and reliable, but is worth half the capacity.
RAID is not a backup!
RAID protects against
Disc breakdowns
But not from the removal of files, viruses, fire, or theft.
03
Cloud storage
Yandex Disk, Google Drive, Dropbox, and others' computers
GEF
What's a cloud?
Cloud Warehouse
The service is a service where data are stored on the provider's servers and available via the Internet. You pay for the volume, the provider provides reliability.
Service models
Model
What do you get?
Examples
IaaS
Virtual servers + disks (infrastructure)
Yandex Cloud, AWS EC2, Selectel
PaaS
Development platform + repository
Google App Engine, Heroku
SaaS
File storage application complete
Yandex Disk, Google Drive, Dropbox, OneDrive
How to Ensure Reliability
Replication
- each copy of the file is stored in several data centres.
Erasure coding
- file split into fragments + excess codes (like RAID 5/6, but for data centres)
Geo-irrigation
— copies in different regions (Moscow, Krasnodar, Siberia)
Relationship to Chapter 4
In chapter 4, you learned about network topology and equipment. Cloud vaults are thousands of servers in data centres connected to the topology of the "composite network" (mesh) with excess channels.
04
Distributiond storage
P2P, blockcase, Hadop — Data without a single center
PIF
Distributed File Systems
Data are broken down into blocks and distributed by
multiple knots
Each block is duplicated on several nodes. If one server fails, the data is taken from the other.
Examples
HDFS
(Hadop Distributed File System) is a store of huge files broken into 128 MB blocks, and each block is duplicated three times.
IPFS
The Internet File System is a decentralized P2P network that is located on the contents (hash) rather than on the location of the network.
Block
- A distributed register where each of the node stores a complete copy of the data.
CAP Theorem Principle
In a distributed system, it is not possible to simultaneously achieve all three properties:
Consistency
(harmonization)
Availability
(accessible)
Partition tolerance
(resistance to separation) You have to choose two out of three.
💾
RAID SIM
RAID 0 · 1 · 5 · 10 · Data distribution
Interactor
Select the RAID level and see how the data is distributed on disks. Press the disk to break it and see if the data survives.