Vous êtes sur la page 1sur 48

1 Patni Internal 1

Storage Technology Primer

2 2
Copyright ©2006 by Patni
Coverage

• Need for Storage & Growth of Data


• Why Network Storage?
• Storage Performance and Data Protection
• DAS, NAS, CAS, SAN
• Connectivity Technology
• Standards Organization

3 3
Copyright ©2006 by Patni
Need for Storage

4 4
Copyright ©2006 by Patni
Growth of Data

World Wide Production of Information


200
180
160
140
Exabytes*

120
100
80
60
40
20
0
1997 1998 1999 2000 2001 2002 2003 2004 2005 2006

*Exabyte = One Billion Gigabytes

5 5
Copyright ©2006 by Patni
Volatile Storage

• Volatile Storage – A storage device in which the contents


are lost when power is removed.
• Cache Memory -As the microprocessor processes data, it
looks first in the cache memory and if it finds the data
there (from a previous reading of data), it does not have
to do the more time- consuming reading of data from
larger memory.
• Disk Cache - A disk cache is a mechanism for improving
the time it takes to read from or write to a hard disk

6 6
Copyright ©2006 by Patni
Non Volatile Storage

Magnetic Optical IC

Disk Tape •CD •FFD


• DVD •SD
•MMC

 ATA Drive •Mammoth


SCSI Drive • AIT
Zip Disk • DLT
USB Disk • LTO

7 7
Copyright ©2006 by Patni
Storage Issues

8 8
Copyright ©2006 by Patni
Storage Issues

9 9
Copyright ©2006 by Patni
Evolution of Storage Management

Enterprise storage Management

70s 80s Late 90s

Host Centric Client/Server Network Computing

10 10
Copyright ©2006 by Patni
Types of Network Storage
• Direct-attached storage (DAS)
Storage devices connected to a server
• Network-attached storage (NAS)
Storage devices can be accessed over a computer network
• Storage area network (SAN)
Storage devices can be accessed through Fibre Channel
• Content Addressable Storage (CAS)
Data stored on multiple intelligent nodes

DAS NAS SAN CAS

Fibre IP
IP Channel
LAN/WAN or IP LAN

11 11
Copyright ©2006 by Patni
DAS – Direct Attached Storage

• Storage hardware that connects to a single


server. Also called: Direct Attached
Storage (DAS). DAS is Local storage and
one-to-one relationship.
• Most of storage (about 70%) is connected
via DAS today, but 60% of storage will be
connected via SAN by 2007(IDC)

12 12
Copyright ©2006 by Patni
DAS (cont’d)

• Advantages •Disadvantages
Low capital expense, The least flexible of all data
industry-standard parts sharing topologies

Less than NAS file serving


Simple straight forward
performance
setup
Large administrative effort and
Can run applications and
perform as a file server
high cost to maintain
Relative high-performance
Minimal data sharing
at low-cost within a given
DAS island

13 13
Copyright ©2006 by Patni
Network Storage Evolution
DAS SAN
clients

LAN

Disk drives

Disk drives Server


Server

NAS

14 14
Copyright ©2006 by Patni
Content Addressed Storage
• Released in 2002 by EMC
• Refers to storage of fixed content, that is data written
once and never changed or cant be altered in any way.
• Storage schema was designed specifically for keeping
fixed content secure and in place was inevitable.
• CAS provides a digital fingerprint for a stored piece of
data. Also known as an ID or logical address.

15 15
Copyright ©2006 by Patni
Requirement of CAS

• Guaranteed integrity and authenticity of information


• Location independent & Long-term retention

• Continuous online and available


• Automatic distribution across devices

• Operation and management of information should be automated

• Singapore context
MAS rule that financial documents must be retained for 6 years
Companies Act, all financial records must be maintained for 7
yearsTelecom need to keep their CDRs for 3 years

16 16
Copyright ©2006 by Patni
How CAS can Impact Business

• Confront the increasingly stringent business and legal requirements to

store information for decades.


• Storage ecosystem that can alleviate one of today's growing storage

challenge.
• Its core software can manage over decades, extraordinary amounts of

fixed content and to integrate easily with Independent software vendor


applications.
• Modification of existing applications via its application programming

interfaces.

17 17
Copyright ©2006 by Patni
How does CAS Work?

• Uses a ‘claim check’ to address objects.


• FCS generates a globally unique identifier (claim
check) to later retrieve the object. The client
presents a token and the server replies with the
object.
• The token is a set of ASCII characters. Maybe fixed in
length or varies. Token is a unique string associated
with each object in the system.

18 18
Copyright ©2006 by Patni
SAN, NAS & CAS- Comparison

Block File Object

SAN NAS CAS

High High
Performance Capacity
Mixed
Performance/Capacity

19 19
Copyright ©2006 by Patni
RAID

• What is RAID?
• Stands for Redundant Array of Independent Disks.
• Developed to meet the growing demands for data reliability and
performance.
• Multiple hard drives are grouped together to form a single
logical drive.
• Why RAID?
• Mass storage is successful only with the benefits of this
data protecting scheme.
• Increases the performance and reliability of data storage
by spreading data across multiple disks.

20 20
Copyright ©2006 by Patni
RAID Concepts
• RAID uses Mirroring, Parity and Striping.

•Mirroring

• Increases fault tolerance by having two copies of the same data on

separate hard drives.


• Downtime is minimal and data recovery is simple.

• Increased cost and twice as much as storage.

Data(ABC) RAID A B C Disk 1


Controller

A B C Disk 2
21 21
Copyright ©2006 by Patni
RAID Concepts (cont’d)

•Striping
•Improves performance by distributing data across all drives.
• The transfer rates for read and write operations are greatly
increased.
• There are two levels of striping:
• Byte level striping: breaking up of data into bytes.
• Block level striping:breaking up of data into specific block sizes.

Data(ABC DEF) A C E Disk 1


RAID
Controller

B D F Disk 2
22 22
Copyright ©2006 by Patni
RAID Concepts (cont’d)

• Parity
Data redundancy technique used in RAID.
Parity data is created using the logical operation
called XOR on the data elements.
If any of the data elements is lost, it is recreated
from the parity element and vice versa.
As in mirroring there is no need to keep two copies of
data.
The parity can be either distributed across the
multiple disks or be dedicated to a single disk.

23 23
Copyright ©2006 by Patni
RAID Levels

• Combinations of mirroring, parity and striping results in

various raid
Commonly levels.
adopted RAID Levels
• RAID 0 – Striping (no parity)
• RAID 1 – Disk mirroring
• RAID 0+1 – Striping, each stripe then mirrored
• RAID 2 – Bit-level Striping, ECC Disk
• RAID 3 – Byte-level Striping, fixed parity
• RAID 4 – Block-level Striping, fixed parity
• RAID 5 – Striping, distributed parity
• RAID 6 – is two parity over all drives, Handles two disk failures.

24 24
Copyright ©2006 by Patni
RAID 0: Striping

• Implements a striped disk array, the data is broken into


blocks – each block written to separate disk drive

• Not fault-tolerant – not a “true” RAID

• Lower cost & higher access rate

• Applications in Image Editing & Video Production

Block11
Block Block22
Block
Block33
Block Block44
Block
Disk 1 Block55
Block Block66
Block Disk 2
Block77
Block Block88
Block
25 25
Copyright ©2006 by Patni
RAID 1 : Mirroring
• Consists exactly 2 disk modules bound together as mirrored pair.
• Controller must perform 2 concurrent Reads or 2 duplicate
Writes, per mirrored pair.

• If both disks fail, the RAID 1 mirrored pair becomes inaccessible

• Recommended Application-Accounting & Payroll.

Block11
Block Block11
Block
Block22
Block Block22
Block
Disk 1 Block33
Block Block33
Block Disk 2
Block44
Block Block44
Block
26 26
Copyright ©2006 by Patni
RAID 0/1 : Striping & Mirroring

• Even number of 4-16 disk modules


• Half are data disks and the other half are disk mirrors
• Uses block striping for performance & mirroring for redundancy
- so mirrored RAID 0 group

Block11
Block Block22
Block Block11
Block Block22
Block
Block33
Block Block44
Block Block33
Block Block44
Block
Block55
Block Block66
Block Block55
Block Block66
Block
Block77
Block Block88
Block Block77
Block Block88
Block
27 27
Copyright ©2006 by Patni
RAID 2 : Bit-level Striping, Fixed parity

• Uses bit level striping with dedicated Error Correction Code


(ECC).
• Use multiple drive dedicated ECC disks.
• Need high number of drivers for ECC generation.

Block11
Block Block1a
Block 1a Block1b
Block 1b ECC
ECC ECC
ECC
Block22
Block Block2a
Block 2a Block2b
Block 2b ECC
ECC ECC
ECC
Block33
Block Block3a
Block 3a Block3b
Block 3b ECC
ECC ECC
ECC
Block44
Block Block4a
Block 4a Block4b
Block 4b ECC
ECC ECC
ECC

28 28
Copyright ©2006 by Patni
RAID 3 : Byte-level Striping, Fixed parity

• Uses byte level striping with dedicated parity

• An additional drive dedicated to parity

• Uses Error Correction Code to detect errors

• Added parity slow down writes

Block11
Block Block1a
Block 1a Block1b
Block 1b parity
parity
Block22
Block Block2a
Block 2a Block2b
Block 2b parity
parity
Block33
Block Block3a
Block 3a Block3b
Block 3b parity
parity
Block44
Block Block4a
Block 4a Block4b
Block 4b parity
parity
29 29
Copyright ©2006 by Patni
RAID 4 : Block-level Striping,Fixed parity

• Uses block level striping with dedicated parity


• Multi user read, single user writes .
•Uses Error Correction Code (ECC) to detect error.
•Added parity slow down writes.

Block11
Block Block22
Block Block33
Block parity1
parity1
Block44
Block Block55
Block Block66
Block parity2
parity2
Block77
Block Block88
Block Block99
Block parity3
parity3
Block10
Block 10 Block11
Block 11 Block12
Block 12 parity4
parity4

30 30
Copyright ©2006 by Patni
RAID 5 Striping and Distributed
Parity
• Uses block level striping with distributed parity

• Data transferred to disk by independent read and write


operations
• Combination of redundancy, cost effectiveness and storage
efficiency
• Good for multitasking environment.

Block11
Block Block22
Block Block33
Block parity
parity
Block1a
Block 1a Block2a
Block 2a parity
parity Block44
Block
Block1b
Block 1b parity
parity Block3a
Block 3a Block4a
Block 4a
parity
parity Block2b
Block 2b Block3b
Block 3b Block4b
Block 4b
31 31
Copyright ©2006 by Patni
RAID 6 Two Parity Over All Drives

• 2 parity (P,Q) Redundancy, over all drives.


• Handles two disk failures

Block11
Block Block22
Block Block33 Parity1a
1a Parity1b
1b
Block Parity Parity
Block44
Block Block55
Block Parity2a
2a Parity2b
2b Block66
Parity Parity Block
Block77
Block Parity3a
Parity 3a Parity3b
3b Block88 Block99
Parity Block Block
Parity4a
Parity 4a Parity4b
Parity 4b Block10
10 Block11 11 Parity5a
5a
Block Block Parity
Parity5b
Parity 5b Block12
Block 12 Block13
13 Parity6a
6a Parity6b
6b
Block Parity Parity

32 32
Copyright ©2006 by Patni
Solutions for Connectivity

• Fibre Channel
• iSCSI (Internet Small Computer System Interface)
• FCIP (Fibre Channel Over Internet Protocol)
• iFCP ( Internet Fibre Channel Protocol )

33 33
Copyright ©2006 by Patni
Introduction to Fibre Channel

• High Speed: 2 Gbps, full duplex dedicated


connection.
• Provides a general hardware transport vehicle for
Upper Level Protocols (e.g. SCSI, IP, etc)
• Reduced congestion
• Long Distance: Up to 10km.
• Heterogeneous systems support

34 34
Copyright ©2006 by Patni
Fibre Channel Architecture

ULP (Upper Level Protocol) SCSI-3 IP ATM

IPI - 3 Command FC Link


FC-4 SCSI - 3 Command
Set Mapping Encapsulation FC - ATM
Set Mapping
(IPI-3 STD) FC - LE

FC-3 Common Services

FC-2 Framing Protocol


FC - AL FC - AL -2
Fibre Channel Physical
FC-1 Encode / Decode & Signaling Interface
8B/10B Encoding
( FC- PH, FC-PH2,
FC-0 Physical Variant FC-PH3 ) Copper, Optical
35 35
Copyright ©2006 by Patni
The Fibre Channel Network

SCSI Initiator Node

SCSI Command Storage


Protocol
Topology Subsystem
Link

SCSI Target Node


Server
IP Node
IP Node
Name=‘1000….3210’
Name=‘1230….3210’

36 36
Copyright ©2006 by Patni
Fibre Channel Topologies

Point To Point Arbitrated Loop Switched Fabric

Only 2 devices Up to 126 Up to 16 million devices

Direct Connect devices FC switches


FC Hubs

37 37
Copyright ©2006 by Patni
Fibre Channel Node and Port
 Node: Any device or entity that can initiate and receive transmissions
in a fibre channel network.Eg. Storage device
 Port: A fibre Channel port is an intelligent physical hardware that
connects all components to the network.
 Switch and hubs are not considered as nodes as they do not initiate
any transmissions.
 A node can have multiple ports Node

Node Name=‘1000…ABCD’ N_Port N_Port Port Name=‘2100…


#1 #2 8765
Nodes N_Port NL_Port N_Port_ID = ‘123456’
#3 #4
Each Port has
separate Transmit and
U_Port U_Port Receive functions

NL_Port NL_Port
#7 #8
38 38
Copyright ©2006 by Patni
Node Name=‘6000..1254
Fibre Channel Addressing
Each Node has a unique Node_Name ( 64-bit WWNN)
assigned by the manufacturer.
 As in LAN MAC address, WWN not used for transportation of
frames across the network.
 Each Node has one or more ports called N_Port (NL_Port)
 Each N_Port has a 64-bit Port Name and a 24-bit port address
or N_Port_ID
 The 24-bit address is assigned dynamically and optimizes
frame routing.
Multi-ported Node
64 bit Node_Name or WWN

64 bit Port Name Node Port or N_Port


24 bit39
N_Port_ID or port address 39
Copyright ©2006 by Patni
Fibre Channel Port Types

 Switch Ports
E_Port = “Expansion Port”
F_Port = “Fabric Port”
FL_Port = “Fabric Loop Port”
G_Port = “Generic Port – Can Operate as E, F or FL”
 Device Ports
N_Port = “Direct Fabric Attached Device”
NL_Port = “Loop Attached Device”

40 40
Copyright ©2006 by Patni
Port Visualization

Switch 1 N-Port Node


Node N-Port F-Port F-Port

E-Port
Fabric
E-Port
NL-Port
FL-Port
G-Port
NL-Port
Switch 2
41 41
Copyright ©2006 by Patni
Brocade
Classes of Service
• High speed and Flexible architecture of Fibre Channel enables
it to carry out different applications.eg tape-backup, real-time
video etc.
• Different applications have their own delivery requirements like
bandwidth, connectivity etc.
• Fibre Channel defines 5 classes of service that address
different delivery requirements
Class 1- Acknowledged Connection Oriented Service
Class 2- Acknowledged Connection-less Service
Class 3 - Unacknowledged Connection-less Service
Class 4 – Fraction Bandwidth Connection Oriented Service
Class 6 – Multicast Service.

42 42
Copyright ©2006 by Patni
Thanks

43Patni Internal 43
Class 1 Service
• Connection-Oriented Class of Service.
• Allows full-bandwidth between a pair of nodes with a
confirmation of delivery and End To End Flow Control
• In-Order delivery of frames due to dedicated connection.
• Best suited for real-time applications like video processing.
• Connections are established and removed at sequence boundaries.

N_Port Fabric N_Port


#1 #3

N_Port N_Port
#2 #4
44 44
Copyright ©2006 by Patni
Class 2 Service
• Connection-less service class and no bandwidth is reserved.
• Frames are routed by the fabric using internal routing algorithms.
• Do not guarantee In-Order delivery of frames, but there is
confirmation of delivery.

N_Port Frame 2:1 N_Port


#1 #2
Frame 1:5
Frame 2:2

Frame 1:5

N_Port N_Port
#2 #5

45 45
Copyright ©2006 by Patni
Class 3 Service

• It is a connection-less and unacknowledged service.


• Frame routing done internally by the fabric.
• Also known as the datagram service, as it is the quickest
transmission without involving the acknowledgement.
• If a frame cannot be processed then, it is discarded without
notification.

First frame of sequence

Sequence Data Frame Sequence


Initiator Recipient
Last data frame
First frame of sequence
Sequence Data Frame Sequence
Recipient Initiator
Last data frame

46 46
Copyright ©2006 by Patni
Class 4 Service
• This service is connection-oriented providing a virtual circuit
between a pair of nodes.
• Known as Fractional Bandwidth, as it assigns only a fraction of
the total bandwidth and different QoS parameters for each
connection.
• The QoS parameters for each services ensures that certain time-
critical applications always have bandwidth available.
• Exists only as a standard due to its complexity.

47 47
Copyright ©2006 by Patni
Class 6 Service
• Provides reliable multicast service with acknowledged delivery.
• It is only a variation from Class1 service in a way the responses
are processed by recipients of the multicast group.
• A multicast server serves consolidates all acknowledgements,
thereby returning only a single confirmation to the initiator.
• Example, A video broadcast application, with a central video
server and multiple video recipients.
ACK’s
from the
Initiator Connect Fabric Recipient targets
request
N_Port N_Port
N_port
Multicast
N_Port
ACK server

48 48
Copyright ©2006 by Patni

Vous aimerez peut-être aussi