Oct. 2009

RAID

2

3

AgendaAgenda

1.RAID

2.RAID

3.RAID

4.4.

5.

6.Q & A

4

RAID Redundant Array of Inexpensive Disks (RAID) --

5

(RAID)

1. 1987 University of California - Berkeley A Case for Redundant Arrays of Inexpensive Disks

2.

6

RAID Technology

HostSystem

Disk Array Controller

951

1062

1173

1284

RAIDRAID Level ZeroLevel Zero

HostSystem

Disk Array Controller

321

654

321

654

RAID Level OneRAID Level One

7

RAID Technology (2)

HostSystem

Disk Array Controller

741

852

963

ParityParityParity

RAID Level Three/FourRAID Level Three/Four

HostSystem

Disk Array Controller

741

Parity52

8Parity

3

96

Parity

RAID Level FiveRAID Level Five

8

RAID Technology (3) P + Q Parity

HostSystem

Disk Array Controller

741

Parity52

QParity

3

8Q

Parity

96Q

RAID Level SixRAID Level Six

9

1. (High Performance)

2.(Data Protection)

3. (High Capacity)

10

RAID ROMB

(RAID on Motherboard)

Internal RAID Card

JBOD (SBOD, EBOD)

External RAID Subsystem JBOD

RAID card

ROMBOsc.Voltage

Regulation PCIe 8x

SFP

USB

DDR2DRAM

Flash DDR2DRAM

PCIe 8xDDR2DRAM

DDR2DRAM

Storage ProcessorDAR

ASP

External RAID

11

Architecture:Monolithic vs. Modular Storage Two classes of external storage platforms:

High-end Monolithic Storage SystemsModular dual-controller based systems

12

64-bit/133MHz

InfortrendASIC

500266

PowerPC750FX 600

DDR266SDRAM

64-bitCPU Bus

64-bit/133MHzMemory Bus

PCI-X Bus FC-4G2 ch to host

SAS-3G8 ch to drives

SAS-3G8 ch to drives/exp.

SATA-3G8 ch for RCC

64-bit/133MHzPCI-X Bus

8-bitX-Bus

Misc.I2CLANFlashNVRAM

Battery

13

Redundancy for RAID Controllers

. . . .

12x Hard Drives

SATA8

channels

SAS8 channels

SAS8 channelsFC-4G

2 channels

Infinibandx 4

SAS8 channels

SAS8 channels

FC-4G2 channels

Infinibandx 4

RC

C link

Expansion Channels

cross-links

Com

mon B

ackplane

RAID Controller A

RAID Controller B

2 x

2 x

2 x

2 x

To JBOD

To JBOD

SFP

SFP

SFP

SFP

8 x

4 x

8 x

4 x

SATA8

channels

2 x

2 x

MUX

MUX

MUX

MUX

MUX

MUX

MUX

MUX

SASrepeater

SASrepeater

FC-4GCH0

FC-4GCH1

FC-4GCH1

FC-4GCH0

Oct. 2009

15

RAID Architecture & Components

Discrete Design vs. SOC Protocol IC (IOC)

Fibre ChannelSerial Attached SCSIiSCSIInfiniteband

RAID Chipset (RAID H/W) Processor IC (IOP) IOP + Target Phy ROC (RAID on Chip)

16

Infortrend ASIC Design More interfaces for I/O bandwidth RAID H/W engine (RAID 5 XOR, RAID 6 P+Q) Advanced Features

PPC CPU

IFT ASIC 667

Memory Redundant Link

Memory

FC-4G2 channels

SAS8 channels

PCI-Express

17

Storage IC Design Functional specification Algorithm implementation Performance balance IP integration SOC verification Net-list Physical design

Power estimationDie sizeI/O and PackageFloor planTiming closure

System verification Electrical characterization Performance analysis

18

X86-based Design for Storage For example, FC-to-SAS ROMB or internal RAID card to drives

X86 CPU

ChipsetMemory Redundant Link FC-4G

2 channelsRAID

PCI-Express

SAS8 channels

19

X86-based Design for Storage (2) For example, FC-to-SAS Independent RAID ASIC coworking with Intel

ArchitectureX86 CPU

Chipset

RAID

ASIC

FC-4G2 channels

SAS8 channels

PCI-Express

Redundant Link Memory

20

IO Processor CPU + Chipset Intel IOP341/342 AMCC

440SP/SPe

21

IO Processor + Targets CPU + Chipset + Target (PHY) Intel IOP348 (Sunrise Lake )

22

SoC or ROC (RAID on Chip) Freescale CSP SoC

23

Additional features by SoC Ethernet (for NAS)

USB

Flash Controller

Audio/Video

Data De-duplication

Data Security

Data Integrity

24

(Data De(Data De--Duplication)Duplication)1.

2. Single instance storage (SIS

3.

25

How De-Duplication Works? Example from IBM

26

Hardware Acceleration for Data DeHardware Acceleration for Data De--DuplicationDuplication

Hardware Computation for Hash functionData de-duplication product manufacturers employ cryptographic hash functions to perform checksum processing because of their ability to produce unique fingerprints from unique data patterns.

Data Compare by hardware

27

Data Security (IEEE1619) Security for storage data at rest AES-based encryption Applied to 512-byte blocks Without data expansion (no additional integrity tag) Parallelizable for high speed HW Standard common format for key backup

Allows for decryption of a disk encrypted by any other vendor

28

Data Integrity DIF (Data Integrity Field) Error Protect

Bit error- I/O interface data path error (H/W error)- Wild store error (H/W and F/W error)

Data displacement error- Buffer allocation error (F/W error), detected by checking

reference tag and LBA

8 bytes DIF for each standard 512 block of dataLogical Block Guard (CRC) 2bytesLogical Block Application Tag 2bytes

FFFFh disables checking of all protection information Not defined by standard

Logical Block Reference Tag 4 bytes

29

DIF Diagram

InitiatorBack-EndAdapter

Memory subsystem

HostInitiatorAdapter

DIF Insert

Verify & Delete

Verify & Forward

Data & DIF written

Read Data & DIF

Verify & Forward

HostInitiatorAdapter

(With DIFcapability)

Verify & Forward

Verify & Forward

Server/Host Subsystem Disk Targets

Data (512 byes)

Data + DIF (512 + 8 = 520 byes)

TargetFront-EndAdapter

Storage Controller

Data Write flowData Read flow

Host 1

Host 2

SCSI / FC / SAS PCI / PCI-X / PCI-Express SCSI / FC / SAS / SATA

30

From Scale Up to Scale OutGrid/Brick/Blade Architecture

- IBM XIV- Panasas

Network RAIDP2P architecture

More tolerance required

Storage Pool

Scale Up Scale Out

31

1. 2. 3.

1. 2. 3.

32

Mechanical Design ID design Drop/Vibraion solution Thermal solution EMI/Safety solution Packing design

Software Design RAID management utilities

design Storage Application

Design(MPIO, Data Service)

RAID Subsystem front view

Server

RAID Subsystem back view

ASIC DesignHW Design System Architecture design Circuit design PCB Layout Signal Integrity solution EMI solution

RAID Controller

Firmware Design (in Flash) RAID Kernel design RAID Processing design Device drivers design

Oct. 2009

Reliable Networked Storage Solution ProviderReliable Networked Storage Solution Provider

34

1993

405

28 3500

(Networking Storage Solution)

TAIWAN / 1993

USA /1996

UK / 2001

CHINA / 2000

JAPAN / 2004

35

///

1. (High Performance)(High Performance)

2.2. (High Capacity)(High Capacity)

3.3.(Data Protection)(Data Protection)

36

57.28%

93 94 95 96 97

3,071,342 2,942,463 3,144,917 3,182,972 2,592,977

7.16 5.99 4.74 6.10 3.99

37

47%30%23%

38

// ()246 / 409 = 60.15%

///

39

..2009 4

2008 A+73

40

ASIC Engineer /FW Engineer /SW Engineer

Product Manager / International Technical Support Engineer

41

Q & AQ & A